Assessing systematic biases in farmers’ local weather change perceptions Authors: Gaurav Arora, Hongli Feng Published: 2024 ABSTRACT Scientific data concerning climate change are critical for designing mitigation and adaptation strategies. Equally important is how stakeholders perceive climate change because perceptions influence decision-making. In this paper, we employ spatially-delineated primary surveys to evaluate weather perception biases among corn and soybean farmers located on western frontier of the U.S. Corn Belt where substantial loss of grassland has been documented. We characterize farmers’ perception biases by measuring the gap between survey-based perception reports for three distinct weather indicators (i.e., temperature, precipitation and drought) and corresponding meteorological evidence. About 70% farmers in our sample misperceive past weather changes. Three-fourths of these misperceiving farmers over-estimate local temperatures and drought frequency and 40% of them under-estimate precipitation trends relative to past records. We further find evidence that farmers’ weather change perceptions are systematically biased in a manner that would justify past land use decisions. Particularly, higher cropping incidence on previously protected grasslands effected more farmers to under-perceive drier conditions and over-perceive wetter conditions. Our investigation of perception biases across distinct weather indicators with a reference to past economic decisions enriches the understanding of climate change perceptions and related policies.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-024-76327-8. Introduction Climate uncertainty is a critical risk factor in agricultural production. Large spatio-temporal datasets and sophisticated statistical models are employed to estimate the distribution of climate variables at regional, national and global scales1. Data-driven decision support systems are a centerpiece of policy designs that are intended to facilitate agricultural adaptation and mitigation amidst climate uncertainty. Meanwhile, forming climate change beliefs can be challenging given that farmers synthesize weather patterns based on local observations with a sparse sampling of extreme events such as floods, droughts and heat waves. Farmers’ beliefs are further impacted by bounded cognitive capacity2,3, exposure to mixed media signals4,5 and complex social, economic and cultural influences6–8. That farmer perceptions may diverge from climate change evidence has been recognized in the literature (e.g.,9–12). Early studies in this literature documented farmer perceptions about future climate change (e.g.,9–11) while recent papers reported perceptions about regional-level weather changes in the past, with a greater focus on developing country contexts in Africa and Asia (e.g.,2,13,14). Climate/weather perceptions are modeled as a function of farmers’ socio-economic and institutional environments to evaluate how such factors inform their attitudes about climate change15,16. That is, whether farmers believe that climate is changing and what they perceive as climate change are understood to depend upon their social attributes such as age and education, economic situation such as income and farm size; and the willingness to mitigate climate impacts. Despite advancements in global and local climate change estimation and the characterization of climate change perception in societies, the evidence remains rather crude for quantifying the gaps between individual perceptions and scientific data. Studies typically report an aggregate percentage measure for describing how climate perceptions compare with scientific evidence. For example, Barnes and Toma11 reported a survey of 540 Scottish dairy farmers where 46% believed that future global temperatures will increase. Similarly, Mahmood et al.13 reported that 70% of 671 farmers in northern Pakistan perceived regional temperature changes in line with the meteorological trends of past 38 years. Systematic patterns in how individual farmers’ perceptions may differ from local weather observation, and the relevant drivers, remain largely unexplored. An understanding of heterogenous weather perceptions can inform climate policy designs by accounting for local climate information as well as farmers’ awareness and motivation to mitigate climate impacts15. Climate change and climate policies are known to generate unequal costs that depend upon local resource constraints and socio-cultural attributes of farmers. On the other hand, localized weather events can influence farmers’ climate beliefs 15–17, through the use of heuristics or psychological manipulations in processing weather information18. In this paper we analyze potential biases in weather change perceptions, which are defined as the gap between farmer perceptions about local weather changes and corresponding data-based evidence for three weather indicators, i.e., temperature, precipitation and droughts. Our analysis seeks to address farmers’ subjective climate change attribution to a particular indicator (or a set of indicators) while many existing studies tend to examine perceptions with reference to “climate change” as a general concept9,10,19,20. We further examine farm-level heterogeneity in weather perception biases with a focus on the role of farmers’ land use decisions in comprehending such biases. It is well-established that climate and weather perceptions influence growers’ attitudes and decisions concerning climate mitigation and adaptation15,16,21,22. For example, Mase et al.23 reported that a decision to purchase insurance or adopt technology to mitigate climate risk was associated with farmers’ climate beliefs. We emphasize that perceptions and decisions can feed into each other, i.e., perceptions affect decision making and decisions can also affect perceptions, in particular, weather perception biases. This latter channel of impacts is afforded much less attention in the literature, although some studies suggest that local experiences and economic activity can affect farmers’ weather perceptions2,24. To our best knowledge, no study has examined whether farming decisions can lead to potentially systematic biases in weather change perceptions. We argue that, in forming weather and climate perceptions, farmers may be motivated to subconsciously distort weather signals25 through social mediation26 and exposure to market or policy shocks8,19 for validating or justifying their past land use decisions. Understanding how past decisions can contribute to perception biases would shed light on how people process climate information in a manner that might differ from objective data. We assess the linkage between farm decisions and weather perceptions in two steps. First, we employ two-way contingency tables to test whether farmers’ weather perception biases are on-average independent of their land use decisions. In the second step, we recognize that perceptions and decisions are endogenous to each other and use economic returns as an instrument to identify the causal link between farm decisions and weather perception biases. Economic returns directly impact land use decisions; on the other hand, we expect that economic returns will affect perceptions through land use changes, and hence provide a valid instrument to estimate the impact of decisions on weather perception biases27, ch.5. Our study makes three main contributions to the existing literature. First, we document farm-level biases in weather change perceptions by assessing spatially-delineated primary survey data on farmers’ perceptions in relation to the local weather records. Alongside weather change perceptions these surveys provide farmers’ land use information on a regional frontier between Northern Great Plains and Western Corn Belt (see Fig. 1a). Our study region witnessed major concerns over rapid loss of native grassland cover to crop production during the past fifteen years28,29, which provides an ideal setting to analyze potentially systematic biases in weather perceptions with a reference to past land use decisions.Fig. 1(a) Spatial distribution of survey responses located to east of the Missouri River. GHCN weather stations (green triangles) provide temperature and precipitation summaries in each respondent’s locality. (b–d) Drought intensity vector maps for August months in 2004, 2006 and 2014, respectively, overlain with our survey locations. Source: The National Drought Mitigation Center (NDMC) in the United States, https://droughtmonitor.unl.edu/About/Permission.aspx. Maps courtesy of NDMC. (a) Spatial distribution of survey responses located to east of the Missouri River. GHCN weather stations (green triangles) provide temperature and precipitation summaries in each respondent’s locality. (b–d) Drought intensity vector maps for August months in 2004, 2006 and 2014, respectively, overlain with our survey locations. Source: The National Drought Mitigation Center (NDMC) in the United States, https://droughtmonitor.unl.edu/About/Permission.aspx. Maps courtesy of NDMC. Second, our inquiry moves beyond the use of a general reference to “climate change” and accounts for regional variations in distinct weather indicators (temperature, precipitation, and droughts) to test the hypothesis that perceptions are consistent with local meteorological data. Spatial heterogeneity in climate data and survey-based weather perceptions allow us to estimate farmers’ weather perception biases, i.e., the difference between perceptions (reported in primary surveys) and meteorological data, conditional on local biophysical and socio-economic contexts. Third, we recognize that perceptions of weather and climate trends can be systematically affected by past decisions, beyond the role of perceptions in affecting farming decisions. We estimate the impact of past land use decisions on farmers’ weather change (mis)perceptions. The theory of “motivated beliefs” provides a potential rationale for how farmers might subconsciously form weather perceptions to justify past decisions30,31. Section "Materials and Methods" describes the data and methods, followed by section "Results". We conclude with a discussion of our main findings in section "Discussions and conclusion". Materials and methods Data and data processing Survey data: farmers’ weather change perceptions and land use decisions We utilize 1026 responses from a survey of farmers located on the east of the Missouri River across 20 counties in North Dakota (ND) and 37 counties in South Dakota (SD), see Fig. 1a. Surveys were conducted from March through May in 2015 among farmers who operated 100 or more acres of land for wheat, maize, soybeans or grass/hay cultivation. Sample observations were distributed across counties in proportion of the eligible farms and the survey response rate was 36%. Farming/ranching was principal occupation of 87% respondents and 94% respondents operated the same farm/ranch for more than 10 years at the time of survey. Also, according to USDA Economic Research Service factsheets32, Dakotas experienced a steady decline in rural population shares during our study period. Thus, it is reasonable to consider our survey respondents to be long-term residents and well-aware of weather conditions in their local area. The survey was designed to primarily study farmers’ cropland and pastureland use decisions during the commodities boom of 2004–2014, and has been utilized to shed light on important land use trends and underlying drivers28,29. Farmers were first inquired about their land use decisions and the potential drivers, followed by perceptions of past changes in local temperature, precipitation and droughts (discussed hereafter). We also collected farmland quality, income status and demographic information. The survey questionnaire pertaining to our study is provided in Supplementary Information (SM), section S1. These data were obtained within the guidelines of the Institutional Review Board of Iowa State University (please see “Availability of data and materials” statement for details). We assess the representativeness of our survey data relative to the 2017 Census of Agriculture. Our sample constitutes farmers with 100 or more farmland acres, a criterion used to screen out very small farms or hobby farmers. With a conservative corn yield of 120 bushels per acre and corn price of $3 per bushel, a 100-acre farm generates at least $36,000 in sales. In contrast, for Census purposes, any plot that generates $1,000 or more in revenues is interpreted as a farm. Our screening criterion will imply some differences between our sample and the Census data. For example, about 25% farms in our sample were 2,000 acres or larger in size, which is well-aligned with the upper quartile of Census farms in the Dakotas. The average farm size (1,650 acres) in our surveys was larger than the average Census farm (1,450 acres). More than 95% farmers in our sample reported $50,000 or more in annual farm sales, which equals the median value of annual sales among Census farms. Farmers’ median age in our surveys, i.e., 50–59 years, was representative of the Census data. A representativeness assessment based on the 2012 Census of Agriculture revealed similar patterns of comparison29. Data summaries are presented in Table 1 and further discussed in the next section.Table 1Summary of survey data attributes for farmers located in the eastern Dakotas.VariableDescriptionNMeanMedianSD\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 1 if ‘cooler’ temperature belief during 2004–2014; = 2 if ‘about the same’; = 3 if ‘hotter’9561.922.000.62\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 1 if ‘lower’ precipitation belief during 2004–2014; = 2 if ‘about the same’; = 3 if ‘higher’9742.092.000.78\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 1 if ‘less droughts’ belief during 2004–2014; = 2 if ‘about the same’; = 3 if ‘more droughts’9611.902.000.71\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}Weather-station record of growing-season temperature (oC) change, 2004–20141026− 0.22− 0.501.65\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document}Weather-station record of growing-season precipitation change (mm), 2004–2014102617.4518.0064.43\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document}USDM drought frequency record, 2004–20141026− 3.43− 4.003.56ageAge category (see notes)10263.303.001.08educEducation levels (see notes)10262.953.000.92incAnnual gross farm/ranch sales (see notes)10263.604.001.53helPercent highly erodible soil10268.311.0015.71heavyPercent heavy soil102637.3330.0032.65latCoordinates along the north–south direction102645.2945.161.29lonCoordinates along the east–west direction1026− 98.0− 98.01.07\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} = 1 if CRP land was converted to cropland; = 0 if not converted during 2004–201410261.702.000.521. Missing data and ‘Don’t Know’ responses were removed for summarizing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}.2. Age categories were assigned as: 19–34 (1); 35–49 (2); 50–59 (3); 60–69 (4); 70 + (5).3. Education levels were assigned as: Highest education level: less than high school (1); high school (2); technical school (3); 4-year college (4); advanced degree e.g. Masters, etc. (5).4. Annual gross farm/ranch sales were categorized as: < $50 K (1); $50 K-$99.9 K (2); $100 K-$249.9 K (3); $250 K-$499.9 K (4); $500 K-$999.9 K (5); $1 million + (6). Table 1Summary of survey data attributes for farmers located in the eastern Dakotas.VariableDescriptionNMeanMedianSD\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 1 if ‘cooler’ temperature belief during 2004–2014; = 2 if ‘about the same’; = 3 if ‘hotter’9561.922.000.62\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 1 if ‘lower’ precipitation belief during 2004–2014; = 2 if ‘about the same’; = 3 if ‘higher’9742.092.000.78\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 1 if ‘less droughts’ belief during 2004–2014; = 2 if ‘about the same’; = 3 if ‘more droughts’9611.902.000.71\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}Weather-station record of growing-season temperature (oC) change, 2004–20141026− 0.22− 0.501.65\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document}Weather-station record of growing-season precipitation change (mm), 2004–2014102617.4518.0064.43\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document}USDM drought frequency record, 2004–20141026− 3.43− 4.003.56ageAge category (see notes)10263.303.001.08educEducation levels (see notes)10262.953.000.92incAnnual gross farm/ranch sales (see notes)10263.604.001.53helPercent highly erodible soil10268.311.0015.71heavyPercent heavy soil102637.3330.0032.65latCoordinates along the north–south direction102645.2945.161.29lonCoordinates along the east–west direction1026− 98.0− 98.01.07\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} = 1 if CRP land was converted to cropland; = 0 if not converted during 2004–201410261.702.000.521. Missing data and ‘Don’t Know’ responses were removed for summarizing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}.2. Age categories were assigned as: 19–34 (1); 35–49 (2); 50–59 (3); 60–69 (4); 70 + (5).3. Education levels were assigned as: Highest education level: less than high school (1); high school (2); technical school (3); 4-year college (4); advanced degree e.g. Masters, etc. (5).4. Annual gross farm/ranch sales were categorized as: < $50 K (1); $50 K-$99.9 K (2); $100 K-$249.9 K (3); $250 K-$499.9 K (4); $500 K-$999.9 K (5); $1 million + (6). Summary of survey data attributes for farmers located in the eastern Dakotas. 1. Missing data and ‘Don’t Know’ responses were removed for summarizing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}. 2. Age categories were assigned as: 19–34 (1); 35–49 (2); 50–59 (3); 60–69 (4); 70 + (5). 3. Education levels were assigned as: Highest education level: less than high school (1); high school (2); technical school (3); 4-year college (4); advanced degree e.g. Masters, etc. (5). 4. Annual gross farm/ranch sales were categorized as: < $50 K (1); $50 K-$99.9 K (2); $100 K-$249.9 K (3); $250 K-$499.9 K (4); $500 K-$999.9 K (5); $1 million + (6). Our surveys recorded farmers’ assessment of change in local weather patterns in 2014 compared to ten years ago. We accommodate the potential inaccuracy or bias in memory recall33–36, through alternative fuzzy delineations of ‘ten years ago’ in meteorological records (details are provided hereafter).That is, respondents were asked to select one among the following options for temperature: ‘warmer’, ‘about the same’, ‘cooler’, ‘don’t know’; precipitation: ‘more’, ‘about the same’, ‘less’, ‘don’t know’; and drought frequency: ‘more’, ‘about the same’, ‘less’, ‘don’t know’. The largest proportion of farmers perceived ‘about the same’ temperature (57% farmers), precipitation (36% farmers) and drought frequency (45% farmers) during 2004–2014. In contrast, 22% and 14% farmers perceived ‘cooler’ and ‘warmer’ temperatures, respectively; 24% and 34% farmers perceived ‘higher’ and ‘lower’ precipitation, respectively; and 29% and 19% farmers perceived ‘less’ and ‘more’ frequent droughts, respectively. About 5% responses were ‘don’t know’ and 4% farmers chose not to respond about past weather perceptions. We evaluate a potential causal link between farmers’ weather change perceptions and their enrolment decision into the Conservation Reserve Program (CRP). The CRP is the largest land retirement program in the United States to protect highly erodible and environmentally sensitive lands37. Farmers who enroll into a CRP contract must retire their croplands for an environment-friendly use for 10–15 years, thereby giving up cropping returns in lieu of a fixed annual transfer called the ‘CRP rental payment’. The Dakotas experienced a dramatic decline in CRP acreage during 2004–201438. For example, South Dakota’s CRP coverage declined from 1.45 million acres in 2004 to 930,000 acres in 2014 when 47% of statewide CRP contracts were due to expire and up for renewal in 2007. Such a large proportion of expiring CRP contracts also implied that many farmers had the opportunity to actively consider their CRP enrolment decisions during our study period. Cultivating on CRP lands typically entails substantial costs, including one-time land preparation cost and variable production costs during each subesquent season. Such costs would only be justified if a farmer expects high crop returns, which in turn depend on favorable weather conditions. Given that weather is uncertain and constitutes complex changes in temperature, precipiation, or both, there is scope for cognitive errors in farmers’ weather change perceptions. We will evaluate whether farmers systematically (mis)perceive local weather after they had also chosen to undertake cropping on lands that were originally protected under the CRP contract. According to the theory of motivated beliefs (explained earlier), justifying a past decision to convert CRP land might be one of the potential factors that influence weather perceptions of farmers, particularly in a manner that favors better returns from conducive weather conditions. Weather data: temperature, precipitation and drought frequency We match our geo-referenced farm survey data to weather station-based temperature and precipitation data, and spatial vector maps of a drought index (discussed hereafter). Daily records of temperature and precipitation levels were acquired from 825 weather stations made available by the Global Historical Climatology Network (GHCN)39. Figure 1a presents the spatial distribution of our survey respondents and GHCN stations in the area. We first assign each survey location a ‘daily temperature level’ and a ‘daily precipitation level’ as the mean value of corresponding records from the nearest three weather stations. Distribution of weather outcomes was very similar between nearest-three and nearest-five weather stations. The average value of daily temperature levels during April-August months was denoted as the ‘growing-season temperature’ whereas the sum of daily precipitation levels during April-August was denoted as the ‘growing-season precipitation’ for each survey location. To assess farmers’ drought perceptions, we utilize vector maps made available by the U.S. Drought Monitor (USDM) on a weekly basis since the year 2000. The USDM is jointly produced by the National Drought Mitigation Centre (University of Nebraska-Lincoln), U.S. Department of Agriculture and the National Oceanic and Atmospheric Administration. These maps divide continental U.S. into six drought intensity categories: ‘exceptional drought’; ‘extreme drought’; ‘severe drought’; ‘moderate drought’; ‘abnormally dry’; and ‘no drought’. In Fig. 1b–d we illustrate drought situation in the Dakotas through USDM vector maps during the August months of 2004, 2006 (i.e., drought year) and 2014. We assign each survey location a comparison of category-wise count of drought incidences during the second week of growing-season month (April-August) in 2004 and 2014. Analysis of farmers’ local weather change perceptions A multivariate model to assess consistency between perceptions and data-based evidence We model the ordinal survey responses for farmers’ local weather perceptions as an ordered logistic regression function40 of data-based weather records conditional on soil attributes and farmers’ age, education and income. The ordinal logistic model is appropriate because survey-based responses represented a clear ordering of how weather outcomes change during 2004–2014 (e.g., from low to high temperature, or from less to more precipitation). Specifically, farmer i’s weather change perceptions, denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i}$$\end{document}, were recorded in an ordered fashion corresponding to weather change data, denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w_{i}$$\end{document}, for three indicators: temperature (t), precipitation (p) and drought frequency (d). As such we define \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i}$$\end{document} as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i} = \left\{ {\begin{array}{*{20}l} {1{\text{ if }}i{\text{ believed }}\Delta w_{i} < 0{\text{, i}}{\text{.e}}{., }w_{i} {\text{ was lower in 2014 relative to 2004 (e}}{\text{.g}}{\text{., less precipitation) }}} \\ {2{\text{ if }}i{\text{ believed }}\Delta w_{i} \approx 0{\text{, i}}{\text{.e}}{., }w_{i} {\text{ was about the same in 2014 relative to 2004 }}} \\ {3{\text{ if }}i{\text{ believed }}\Delta w_{i} > 0{\text{, i}}{\text{.e}}{., }w_{i} {\text{ was higher in 2014 relative to 2004 (e}}{\text{.g}}{\text{., more precipitation)}}{.}} \\ \end{array} } \right.$$\end{document} We describe data-based weather records as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w_{i} \in \{ \Delta t_{i}^{04 - 14} ,\Delta p_{i}^{04 - 14} ,\Delta d_{i}^{04 - 14} \}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t_{i}^{04 - 14} \equiv$$\end{document} average growing-season temperature in 2014 minus its 2004 counterpart; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{i}^{04 - 14} \equiv$$\end{document} total growing-season precipitation in 2014 minus its 2004 counterpart; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{i}^{04 - 14} \equiv$$\end{document} growing-season drought intensity in 2014 minus its 2004 counterpart. See SM, section S4 for details on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{i}^{04 - 14}$$\end{document} variable. We recognize that farmer perceptions may represent a comparison of weather conditions in 2014 with their counterparts circa 2004 (rather than precisely for 2004). Thus, in our analysis, we consider alternative years, i.e., 2003–2005, to represent the start year in 2004. The ordered logit model for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i}$$\end{document} is specified as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i} \le 1:\log \left( {\frac{{\theta_{i,1} }}{{\theta_{i,2} + \theta_{i,3} }}} \right) = \gamma_{1} + \beta_{w} \Delta w_{i} + \vec{x}_{i} \vec{\eta } + \varepsilon_{1i}$$\end{document}1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i} \le 2:\log \left( {\frac{{\theta_{i,1} + \theta_{i,2} }}{{\theta_{i,3} }}} \right) = \gamma_{2} + \beta_{w} \Delta w_{i} + \vec{x}_{i} \vec{\eta } + \varepsilon_{2i} ,$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta_{i,j} = \Pr (\Delta \tilde{w}_{i} = j)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j = 1,2,3$$\end{document}. Model (1) specifies the proportional odds of perception responses: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i} = 1$$\end{document} relative to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i} > 1$$\end{document}; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i} \le 2$$\end{document} relative to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i} =$$\end{document} 3; as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w_{i}$$\end{document} conditional on a control vector, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec{x}_{i}$$\end{document}, of size 1 × k. Slope parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta_{w}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec{\eta }_{{k{\text{x}}1}}$$\end{document} are assumed to be the same regardless of the partition in perception responses, which is known as the proportional odds assumption40. Parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma_{2}$$\end{document} characterize the average log-odds of response levels \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w} = 1$$\end{document} (e.g., lower precipitation relative to higher or about the same values) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w} \in \{ 1,2\}$$\end{document} (e.g., lower or about the same precipitation relative to a higher value), respectively. Note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta_{w} < 0$$\end{document} will signify a positive correlation between weather change records and farmers’ perceptions because \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w$$\end{document} measures increase in w during 2004–2014 whereas model (1) specifies log-odds of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}$$\end{document} from higher to lower w values. Finally, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec{\eta }_{{k{\text{x}}1}}$$\end{document} will provide an estimate of the shift in log-odds of perceiving higher to lower w values upon marginal increase in respective control variables. We intend to infer whether farmer perceptions are on average consistent with the corresponding data-based evidence. Below we formalize our hypothesis. Hypothesis I (Weather perceptions versus weather records): Farmer perceptions are on average consistent with the corresponding data-based evidence conditional on farmers’ socio-economic endowments. In the setup of model (1), we formalize this hypothesis as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{0}^{(1)} :\beta_{w} < 0$$\end{document} against the alternative hypothesis that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta_{w}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document} 0. Our surveys reported how farmers perceived differences between weather instances at two distinct time points where the underlying subjective distribution of weather remains unobserved at each time point. We are unable to control for sub-regional or community-scale environmental or other factors that could potentially contribute to sample-based variation in weather change perceptions. Such information gaps due to the use of cross-sectional weather perceptions data can result in inconsistent estimation of model (1). Nevertheless, considering that our surveys were conducted to represent a region in the U.S. upper Midwest with largely homogenous production systems, we conjecture the potential for such biases is limited41. Articulating biases in past weather change perceptions We cross-tabulate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w_{i}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}_{i}$$\end{document} to characterize the distribution of biased and consistent weather change perceptions of farmers. In particular, we categorize past temperature change records as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t_{i}^{04 - 14} \le - 5\%$$\end{document}; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t_{i}^{04 - 14} \in$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$( - 5\% ,5\% )$$\end{document}; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t_{i}^{04 - 14} \ge - 5\%$$\end{document}, to match survey response categories, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}_{i} \in \{ 1,2,3\}$$\end{document}, i.e., cooler; about the same; and warmer temperature in 2014 relative to 2004, repectively. Similarly, categories \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{i}^{04 - 14}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\in \{ \le - 5\% ,$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$( - 5\% ,5\% ),$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge 5\% \}$$\end{document} were matched to response categories, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}_{i} \in \{ 1,2,3\}$$\end{document}, i.e., lower; about the same; and higher precipitation, repectively; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{i}^{04 - 14} \in$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ < 0, = 0,$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$> 0\}$$\end{document} matched to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}_{i} \in$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ 1,2,3\}$$\end{document}, i.e., lower; about the same; and higher drought frequency during 2004–2014, repectively. Note that the threshold level is empirical in nature and meant to facilitate quantitative comparisons between discrete perception categories and corresponding continuous weather records. We provide a logic for our threshold choice, and a check of robustness by matching ‘about the same’ response category to ± 10% and ± 20% changes in temperature and precipitation levels. Our sensitivity analyses yielded very similar results across the use of various thresholds (discussed hereafter in section "Results"). These cross-tabulations, as demonstrated in Table 4 (and discussed in section "Results"), provide a count of farmers in pairwise categories of survey-based weather change perceptions and the corresponding evidence from weather station data. The diagonal elements provide count of farmers in our sample having weather beliefs that are consistent with weather data while the off-diagonal elements refer to those farmers whose weather perceptions were inconsistent or biased when compared with data-based evidence. Note that perceptions may be consistent or biased even among farmers who experienced similar weather conditions in an area. Linking past weather change (mis)perceptions to land use decisions We motivate the linkage between weather perceptions and land use by visualizing spatial patterns of consistent (triangle markers) and biased (star markers) perceptions of droughts and precipitation in Fig. 2 (while omitting ‘about the same’ and ‘don’t know’ responses). We observe that consistent beliefs about higher regional precipitation and less frequent droughts were clustered in proximity of the Missouri River. However, east of the River we notice a tendency to misperceive ‘drier’ weather to be ‘wetter’ in 2014 relative to 2004 (please see SM, section S5 for details). Regional farmlands also occurred as spatially-contiguous tracts such that grasslands located on the east of the River were subject to large-scale transitions for corn and soybean cultivation during 2004–201442. These observations prompt an investigation into the relationship between land use decisions and weather change perceptions.Fig. 2A spatial map of perceptions of past weather for Dakotas’ farmers showing consistent beliefs (triangle markers) and biased beliefs (star markers) of change in (a) drought frequency and (b) precipitation during 2004–2014. Please see SM (section S5) for more details and discussions related to spatial patterns in farmers’ weather change perceptions. A spatial map of perceptions of past weather for Dakotas’ farmers showing consistent beliefs (triangle markers) and biased beliefs (star markers) of change in (a) drought frequency and (b) precipitation during 2004–2014. Please see SM (section S5) for more details and discussions related to spatial patterns in farmers’ weather change perceptions. We intend to inquire whether past land use decisions could lead to a certain bias in farmers’ weather perceptions. That is, we ask whether farmers in our sample subconsciously ignored adverse weather information (e.g., drought) or were over-attentive towards favorable weather (e.g., higher precipitation) to validate a decision to undertake water-intensive cultivation on pre-existing grasslands? The motivated cognition theory of Benabou and Tirole30 can provide one potential economic explanation for such weather misperceptions among farmers. A farmer who converts native grassland for crop production would be motivated to expect favorable crop-based returns through beliefs that prevail over the risk of droughts in her locality. On the other hand, in case the farmer decides not to convert her grassland then she would seek to validate this decision through beliefs that support frequent drought incidence, and hence low returns from cropping. Below we formalize our hypothesis. Hypothesis II (Weather perceptions and motivated beliefs): Farmers’ land use decisions affect their weather change perceptions. In other words, past land use decisions of farmers were systematically related to the biases in their weather change perceptions. To test Hypothesis II, we first conduct a Chi-squared (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi^{2}$$\end{document}) test of independence between farmers i’s perceptions of data-based weather change records and their land use decisions43, ch. 2). Second, we employ an instrumental variable (IV) approach to evaluate the potential for a causal link between farmers’ weather change perceptions and land use decisions. We denote consistent or biased perceptions as part of i’s beliefs about low’ w value in 2014 relative to 2004, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{H} \in \{ L \to H,L \to L\}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to H$$\end{document} indicates misperception of ‘low’ w to be a ‘high’ value, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to L$$\end{document} indicates that i accurately perceived ‘low’ w value. Similarly, we denote \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{H} \in \{ H \to L,$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to H\}$$\end{document} such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to L$$\end{document} indicates that ‘high’ w value was misperceived to be ‘low’ value, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to H$$\end{document} indicates a consistent belief about ‘high’ w in 2014 relative to 2004. Note that we do not consider ‘about the same’ and ‘don’t know’ responses here. As noted earlier, we focus on the decision concerning CRP enrollment (denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}). Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 denote i's decision to convert her CRP land for cropping; whereas \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 denotes either as re-enrollment decision of the expired CRP land or new enrollment into CRP. We define the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi^{2}$$\end{document} statistic based on a cross-tabulation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} and b in a 2 × 2 contingency table (please see Table 5a, b, discussed hereafter in the next section). Each cell in this table provides sample-based count of farmers, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{l,b}$$\end{document}, for decision-belief pairs \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(l,b) \in \{ \Delta l\} \times \{ b\}$$\end{document}. Corresponding to each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{l,b}$$\end{document} value we define an ‘expected’ count of farmers, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{n}_{l,b}$$\end{document}, under the assumption that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} and b are independent variables. The Chi-squared statistic for 2 × 2 contingency tables is defined as having one degree of freedom (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi_{1}^{2}$$\end{document}). That is,2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X\triangleq \sum\limits_{l\in \{0,1\}}{\sum\limits_{\begin{smallmatrix} b\in \{L\to H,L\to L\} \\ \text{or }\{H\to L,H\to H\} \end{smallmatrix}}{\frac{{{({{n}_{lb}}-{{{\hat{n}}}_{lb}})}^{2}}}{{{{\hat{n}}}_{lb}}}}}\sim \chi _{1}^{2}$$\end{document} If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X > \chi_{1*,5\% }^{2}$$\end{document} (or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\le \chi_{1*,5\% }^{2}$$\end{document}) then we will reject (or fail to reject) that land use decisions are independent of farmers’ weather perceptions with 95% confidence (please see SM, section S6). Assessing potential causal impact channel between land use decisions and weather perception biases The contingency tables do not provide sufficient evidence that past CRP land conversion decisions can lead to systematic biases in farmers’ weather perceptions. Weather perceptions and land use decisions may be endogenous to each other due to reverse causality or unobservable confounders, such as farming ability, that simultaneously drive these variables. For example, farmers’ adaptive or coping capacity can influence how they perceive local weather changes (e.g.,3,23). Favorable farming abilities can also effect a costly decision to pursue crop agriculture on CRP lands. We propose to address endogeneity by specifying instruments that directly impact the CRP conversion decision but influence weather perceptions only through the land use channel27, ch. 5)]. Our choice of instruments is motivated by the local context of declining CRP acreage during 2004–2014, which coincided with sharp rise in crop prices and cropland returns44,45. Land-based returns are known to drive land use decisions, and whether to cultivate CRP land or keep it out of production crucially depends on the net returns that represent relative competitiveness of these land use choices. Based on this knowledge and data availability, we consider two IVs. First, we include CRP rental payments to account for annual returns from CRP lands. Higher CRP rent would generate disincentives for conversion. Our second instrument is the ratio of per-acre returns from croplands to those generated from grasslands which tend to have per-acre returns that are closer to CRP rental payments. Thus, this ratio can reflect relative competitiveness between croplands and CRP lands. We will provide a logic for the validity of IVs based on Wooldridge27, followed by tests for overidentifying restriction and weak IVs, with reference to a regression framework that we outline below. We specify a two-stage least squares (2SLS) regression framework to identify the impact of CRP land conversion decisions on the biases in farmers’ weather perceptions. Figure 3 presents the schematic of our regression analysis. We model the CRP land conversion decision, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}, as3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ln \left[ {\frac{{\Pr (\Delta l_{i} = 1)}}{{\Pr (\Delta l_{i} = 0)}}} \right] = \alpha_{o} + \alpha_{1} rent_{i} + \alpha_{2} ratio_{i} + + \vec{z}_{i} \vec{\delta } + \varepsilon_{i}$$\end{document}Fig. 3Schema for IV-based 2SLS regression framework to identify the impact of CRP land conversion decisions on biases in famers’ weather change perceptions. Schema for IV-based 2SLS regression framework to identify the impact of CRP land conversion decisions on biases in famers’ weather change perceptions. In Eq. (3) we model the log-odds of each farmers i's decision \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 against \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 as a function of two IVs: (a) change in per-acre CRP rental payments during 2004–2014, renti; and (b) ratio between 2004–2024 changes in per-acre returns from croplands and grasslands, ratioi, while controlling for land quality and farmers’ socio-economic information in vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec{z}_{i} = \{ hel_{i} ,age_{i} ,inc_{i} \}$$\end{document} (see Table 1). Please see SM, section S7 for details on construction of land value variables: renti and ratioi, pertaining to each survey location. We predicted the CRP conversion probability, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{p}_{i}^{(CRP \to crop)}$$\end{document}, from Eq. (3) to further estimate a model for weather perception biases that we specify as follows,4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ln \left[ {\frac{{\Pr (b_{i}^{L} = \{ L \to H\} )}}{{\Pr (b_{i}^{L} = \{ L \to L\} )}}} \right] = \omega_{0} + \omega_{1} \hat{p}_{i}^{(CRP \to crop)} + \vec{c}_{i} \vec{\tau } + \zeta_{i}$$\end{document}5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ln \left[ {\frac{{\Pr (b_{j}^{L} = \{ H \to L\} )}}{{\Pr (b_{j}^{L} = \{ H \to H\} )}}} \right] = \iota_{0} + \iota_{1} \hat{p}_{j}^{(CRP \to crop)} + \vec{c}_{i} \vec{\kappa } + \xi_{j} ,j \ne i$$\end{document} In Eq. (4) we model log-odds of i's misperceptions against consistent perceptions of ‘low’ w value in 2014 (relative to 2004) as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{p}_{i}^{(CRP \to crop)}$$\end{document} and demographic controls, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vec{c}_{i} = \{ age_{i} ,$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$educ_{i} ,inc_{i} \}$$\end{document}. Similarly, in Eq. (5) we model log-odds of misperceptions against consistent perceptions of ‘high’ w value in 2014. Given that each farmer could only over-perceive or under-perceive change, not both, we have i ≠ j. The IV-conditioned CRP land conversion probability identifies a causal channel between land use decisions and motivated biases in weather perceptions. Overall, coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\iota_{1}$$\end{document} will provide an inference on motivated weather beliefs due to farmers’ past land use decisions. Evaluating the validity of instruments Wooldridge27 provides necessary conditions for valid instruments that we adapt to the context of our study. First, the instruments must be correlated with the endogenous land use decision conditional on controls in Eq. (3). A strong correlation is desired between the instrument and land use decision to avoid misleading inference from models (4)-(5). Second, the instruments cannot be correlated with the error terms in models (4)-(5), failing which the instrument will face the same endogeneity issue as the CRP land conversion decision. Given that under CRP contracts farmers typically engage in grassland-based activities, our two IVs, namely renti and ratioi, are intended to capture the economic drivers of the choice of cultivating crops on the CRP lands against that of staying in the CRP contract. Further, it is reasonable to assume that these IVs do not directly impact weather perceptions but they can potentially generate incentives for land use conversion in the presence of cropping risks due to weather uncertainty. Therefore, we argue that the aggregate or county-level CRP rental rates and the ratio of cropland returns to grassland returns are valid instruments. Results Results are organized around the testing of Hypotheses I and II, and our findings will be interpreted in the context of our study region’s weather conditions and survey-based socio-economic information of farmers. We present variable descriptions and data summaries in Table 1. According to the meteorological evidence as shown in Table 1, farmers in our study region experienced moderate changes in temperature (i.e., 0.2 °C lower in 2014 from an average 17 °C in 2004) and precipitation (i.e., 17.5 mm higher in 2014 from an average 380 mm in 2004). However, there were substantial variations in weather changes across our study region (please see the standard deviation estimates for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document} in Table 1), which allows us to characterize farm-level heterogeneity in weather perceptions biases. Further, 30% farmers in our sample opted out of the CRP contracts for crop production during 2004–2014. We will assess whether such decisions are systematically related to farmers’ weather perception biases. Biases in farmers’ weather change perceptions Regression based assessment of weather change beliefs relative to data-based records [Testing of Hypothesis I] Table 2 presents the results of our multivariate analysis of weather change perceptions in model (1), where we intend to test whether farmers’ perception reports are on average consistent with data-based evidence conditional on socio-economic and geographic information. We find drought records \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{i}^{04 - 14}$$\end{document} to be positively correlated with farmer perceptions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}_{i}$$\end{document} in a statistically significant manner; but an additional drought only led to 4% (significant at α = 0.05) increase in the odds of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}_{i} = 3$$\end{document} relative to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}_{i} \le 2$$\end{document} responses. However, data-based records for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t_{i}^{04 - 14}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{i}^{04 - 14}$$\end{document} are not statistically significantly associated with perceptions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}_{i}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}_{i}$$\end{document}, respectively. Therefore, we conclude that our data analysis provides very limited support for Hypothesis I.Table 2Ordered logistic regressions for farmer perceptions, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}$$\end{document}, as function of local weather change records, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w$$\end{document}, conditional on farm location, and economic and demographic information.VariableModel 1Dependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}Model 2Dependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}Model 3Dependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}Weather change records\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}− 0.01(0.07)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document}− 0.001(0.001)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document}− 0.04**(0.02)Controls (Farm and farmer characteristics)age− 0.11(0.08)− 0.21***(0.07)0.25***(0.06)inc0.03(0.05)− 0.12**(0.05)0.12**(0.04)heavy soils0.001(0.003)− 0.002(0.003)0.004**(0.002)lat− 0.03(0.07)− 0.29***(0.06)0.31***(0.07)-2LogL110215771804Count of Ordered Outcomes\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 1: 222\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 1: 250\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 1: 293\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 2: 585\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 2: 363\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 2: 459\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 3: 144\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 3: 338\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 3: 199N951***p < 0.01; **p < 0.05; *p < 0.10.Here we include only statistically significant results and keep the sample size same across weather indicators to facilitate a comparative analysis. Tables S1–S3, SM provide full results. Table 2Ordered logistic regressions for farmer perceptions, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}$$\end{document}, as function of local weather change records, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w$$\end{document}, conditional on farm location, and economic and demographic information.VariableModel 1Dependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}Model 2Dependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}Model 3Dependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}Weather change records\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}− 0.01(0.07)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document}− 0.001(0.001)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document}− 0.04**(0.02)Controls (Farm and farmer characteristics)age− 0.11(0.08)− 0.21***(0.07)0.25***(0.06)inc0.03(0.05)− 0.12**(0.05)0.12**(0.04)heavy soils0.001(0.003)− 0.002(0.003)0.004**(0.002)lat− 0.03(0.07)− 0.29***(0.06)0.31***(0.07)-2LogL110215771804Count of Ordered Outcomes\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 1: 222\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 1: 250\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 1: 293\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 2: 585\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 2: 363\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 2: 459\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 3: 144\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 3: 338\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 3: 199N951***p < 0.01; **p < 0.05; *p < 0.10.Here we include only statistically significant results and keep the sample size same across weather indicators to facilitate a comparative analysis. Tables S1–S3, SM provide full results. Ordered logistic regressions for farmer perceptions, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}$$\end{document}, as function of local weather change records, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w$$\end{document}, conditional on farm location, and economic and demographic information. − 0.01 (0.07) − 0.001 (0.001) − 0.04** (0.02) − 0.11 (0.08) − 0.21*** (0.07) 0.25*** (0.06) 0.03 (0.05) − 0.12** (0.05) 0.12** (0.04) 0.001 (0.003) − 0.002 (0.003) 0.004** (0.002) − 0.03 (0.07) − 0.29*** (0.06) 0.31*** (0.07) ***p < 0.01; **p < 0.05; *p < 0.10. Here we include only statistically significant results and keep the sample size same across weather indicators to facilitate a comparative analysis. Tables S1–S3, SM provide full results. To further understand farmer perceptions we assess robustness of our regression results to different cognitive factors such as fuzzy memory and recency saliency, as discussed earlier in section "A multivariate model to assess consistency between perceptions and data-based evidence". We use auxiliary data to represent in local weather changes in the past (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w_{i}$$\end{document}), i.e., (i) recent events, i.e., 2006 droughts and 2011 floods; and (ii) intra-seasonal weather changes. Specifically, we evaluate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} in relation to “high” (> 15 mm) precipitation days during 2004–2014, denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{ > 15mm}^{04 - 14}$$\end{document} (where the threshold of 15 mm exceeded 95th percentile of daily precipitation during 2004–2014); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} in relation to late-season (August) droughts, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{August}^{04 - 14}$$\end{document}; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} in relation to the growing-degree days (GDs), i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta GD^{04 - 14}$$\end{document}. GDs represent a non-linear transformation of daily temperature records to measure the beneficial heat exposure for crop production (see SM, section S8 and46 for details). We also evaluate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} relative to early-season (April-June) GDs, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta GD_{April - June}^{04 - 14}$$\end{document}. Intra-seasonal dissimilarities in temperature impacts on farm outcomes have been previously reported (e.g.,47,48). Estimation results are presented in Table 3.Table 3Ordered logistic regressions for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}$$\end{document} as function of auxiliary weather change records i.e., early-season GDs for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}; high rain events (> 15 mm); and late-season droughts.VariableModel 1ADependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}Model 2ADependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}Model 3ADependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta GD_{April - June}^{04 - 14}$$\end{document}0.18*(0.10)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta GD_{April - June}^{06 - 14}$$\end{document}0.17**(0.09)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{ > 15mm}^{06 - 14}$$\end{document}− 0.03*(0.02)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{ > 15mm}^{11 - 14}$$\end{document}− 0.02*(0.01)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{August}^{04 - 14}$$\end{document}− 0.30***(0.10)N951951951***p < 0.01; **p < 0.05; *p < 0.10.Tables S1–S3 (SM) provide full results for these auxiliary models. Table 3Ordered logistic regressions for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}$$\end{document} as function of auxiliary weather change records i.e., early-season GDs for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}; high rain events (> 15 mm); and late-season droughts.VariableModel 1ADependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}Model 2ADependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}Model 3ADependent Var.:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta GD_{April - June}^{04 - 14}$$\end{document}0.18*(0.10)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta GD_{April - June}^{06 - 14}$$\end{document}0.17**(0.09)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{ > 15mm}^{06 - 14}$$\end{document}− 0.03*(0.02)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p_{ > 15mm}^{11 - 14}$$\end{document}− 0.02*(0.01)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{August}^{04 - 14}$$\end{document}− 0.30***(0.10)N951951951***p < 0.01; **p < 0.05; *p < 0.10.Tables S1–S3 (SM) provide full results for these auxiliary models. Ordered logistic regressions for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{w}$$\end{document} as function of auxiliary weather change records i.e., early-season GDs for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}; high rain events (> 15 mm); and late-season droughts. 0.18* (0.10) 0.17** (0.09) − 0.03* (0.02) − 0.02* (0.01) − 0.30*** (0.10) ***p < 0.01; **p < 0.05; *p < 0.10. Tables S1–S3 (SM) provide full results for these auxiliary models. We find that high-rainfall events in more recent years, i.e., 2006 (drought year) and 2011 (flood year), are significantly corrrelated with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}, unlike the high-rainfall events in 2004 that are uncorrelated with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}. This likely reflects recency effects in farmers’ precipitation perceptions, which may influence their farmland use and other production and management decisions (e.g., insurance uptake decision as shown in49). Further, late-season droughts and early-season GDs exhbit a stronger association with perceptions than their respective season-wide counterparts. That is, an additional August drought (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d_{August}^{04 - 14}$$\end{document}) leads to 28% (α = 0.05) higher odds of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}_{i} = 3$$\end{document} relative to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}_{i} \le 2$$\end{document}, and a unit increase in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta GD_{April - June}^{04 - 14}$$\end{document} leads to 18% (α = 0.10) higher odds of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}_{i} = 1$$\end{document} relative to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}_{i} \ge 2$$\end{document}. GDs are a better representation of heat impacts on crop physiology relative to linear temperatures46. Moreover, farmers may be able to adapt to early-season droughts through delayed sowing50 or resistant seeds28 whereas late-season droughts pose a greater risk of loss. The above results suggest that farmers tend to perceive local weather indicators with regards to their impact on crop production (rather than the simple data summaries of growing-season weather). Liverpool-Tasie et al.2 also suggested that climate perceptions could be related to the degree of exposure and potential for impacts on farmers’ economic activities. However, we focus on how these perceptions compare with the data-based evidence for distinct weather indicators, which has received little attention in the existing studies. A recognition that farmer perceptions may differ across weather indicators, and with a reference to farm production activities, will help enrich climate change perspectives and climate policy debates. Lastly, we find geographic and socioeconomic controls to be distinctly associated with farmer perceptions. Older farmers endowed with higher farm revenues and poor-drainage soils are less likely to perceive low precipitation and frequent droughts, which may reflect on these farmers’ resilience to adverse weather as derived from demographic and biophysical attributes. However, these controls are not correlated with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}, i.e., perceived weather changes (Table 2). Mismatch between farmer perceptions and weather data—evidence of potentially systematic biases [Further evidence for Hypothesis I] We build on the regression-based weak correlation estimates between perceptions and weather data to further tease out systematic patterns in farm-level weather perception biases in light of socioeconomic information and on-farm decisions. Specifically, we cross-tabulate weather change perceptions with corresponding data-based evidence during 2004–2014 in Table 4a–c; where the diagonal elements represent beliefs that were consistent with weather data and the off-diagonal elements provide a count of biased beliefs in our sample. The ‘don’t know’ and missing responses for survey-based perceptions are excluded from this part of analysis.Table 4Cross-tabulations between survey-based weather change perceptions and growing-season (April-August) weather change records during 2004–2014.Farmer beliefs about past temperature change:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}Weather station-based temperature change records:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14} \le - 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14} \in ( - 5\% ,5\% )$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14} \ge 5\%$$\end{document}Total\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 1: Cooler weather1633921223\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 2: About the same42510854587\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 3: Warmer weather1062812146Don’t Know409453Missing data132217Total747186931026(a)Farmer beliefs about past precipitation change: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}Weather station-based precipitation change records: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \le - 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \in ( - 5\% ,5\% )$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \ge 5\%$$\end{document}Total\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 1: Less precipitation1065395254\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 2: About the same12775172374\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 3: More precipitation12681139346Don’t Know1151632Missing data221620Total3722164381026(b)Farmer beliefs about drought frequency change:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}Change in recorded drought intensity scores:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14} < 0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14} = 0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14} > 0$$\end{document}Total\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 1: Less drought2401737294\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 2: About the same3385871467\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 3: More drought1243739200Don’t know344745Missing data171220Total7531171561026(c)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w^{04 - 14} \le - 5\% \equiv \frac{{w_{2014} - w_{2004} }}{{w_{2004} }} \le - 0.05;|\Delta w^{04 - 14} | \le 5\% \equiv |\frac{{w_{2014} - w_{2004} }}{{w_{2004} }}| < 0.05; \, \Delta w^{04 - 14} \ge 5\% \equiv$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{w_{2014} - w_{2004} }}{{v_{2004} }} \ge 0.05$$\end{document}, where w represents growing-season temperature and precipitation, respectively. Our results remain quite similar upon choosing a 10% or 20% thresholds. Table 4Cross-tabulations between survey-based weather change perceptions and growing-season (April-August) weather change records during 2004–2014.Farmer beliefs about past temperature change:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document}Weather station-based temperature change records:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14} \le - 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14} \in ( - 5\% ,5\% )$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14} \ge 5\%$$\end{document}Total\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 1: Cooler weather1633921223\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 2: About the same42510854587\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{t}$$\end{document} = 3: Warmer weather1062812146Don’t Know409453Missing data132217Total747186931026(a)Farmer beliefs about past precipitation change: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document}Weather station-based precipitation change records: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \le - 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \in ( - 5\% ,5\% )$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \ge 5\%$$\end{document}Total\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 1: Less precipitation1065395254\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 2: About the same12775172374\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{p}$$\end{document} = 3: More precipitation12681139346Don’t Know1151632Missing data221620Total3722164381026(b)Farmer beliefs about drought frequency change:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document}Change in recorded drought intensity scores:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14} < 0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14} = 0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14} > 0$$\end{document}Total\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 1: Less drought2401737294\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 2: About the same3385871467\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tilde{d}$$\end{document} = 3: More drought1243739200Don’t know344745Missing data171220Total7531171561026(c)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w^{04 - 14} \le - 5\% \equiv \frac{{w_{2014} - w_{2004} }}{{w_{2004} }} \le - 0.05;|\Delta w^{04 - 14} | \le 5\% \equiv |\frac{{w_{2014} - w_{2004} }}{{w_{2004} }}| < 0.05; \, \Delta w^{04 - 14} \ge 5\% \equiv$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{w_{2014} - w_{2004} }}{{v_{2004} }} \ge 0.05$$\end{document}, where w represents growing-season temperature and precipitation, respectively. Our results remain quite similar upon choosing a 10% or 20% thresholds. Cross-tabulations between survey-based weather change perceptions and growing-season (April-August) weather change records during 2004–2014. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w^{04 - 14} \le - 5\% \equiv \frac{{w_{2014} - w_{2004} }}{{w_{2004} }} \le - 0.05;|\Delta w^{04 - 14} | \le 5\% \equiv |\frac{{w_{2014} - w_{2004} }}{{w_{2004} }}| < 0.05; \, \Delta w^{04 - 14} \ge 5\% \equiv$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{w_{2014} - w_{2004} }}{{v_{2004} }} \ge 0.05$$\end{document}, where w represents growing-season temperature and precipitation, respectively. Our results remain quite similar upon choosing a 10% or 20% thresholds. About 28% (283 out of 1,026) growers report beliefs that were consistent with temperature changes during 2004–2014 while such beliefs were biased for 673 farmers. Within the latter group, 531 farmers misperceived ‘lower’ temperature in 2014 to be either ‘about the same’ or ‘higher’ relative to the 2004 levels. Further, about 30% farmers’ perceptions agreed with past precipitation data whereas 666 farmers misperceived these changes (Table 4b). Among these 666 farmers 267 of them misjudged ‘higher’ precipitation change to be either ‘lower’ or ‘about the same’; 253 farmers misjudged ‘lower’ precipitation to be either ‘lower’ or ‘about the same’; and 134 farmers misjudged ‘about the same’ precipitation to be either ‘lower’ or ‘higher’ relative to 2004 levels. Similarly, 33% growers exhibited consistent beliefs about drought frequency while 622 farmers misperceived them; and among them 462 farmers misperceived less frequent droughts to be ‘more frequent’ or ‘about the same’ (Table 4c). In summary, the largest proportion of farmers misperceived past weather changes, which is in contrast with Hypothesis I. More importantly, Dakotas’ farmers predominantly over-estimated temperature and drought levels, and under-estimated precipitation levels relative to data-based records. We want to note that observing some biases in farmers’ weather perceptions is not surprising given that it is a cognitively challenging task to summarize weather trends that can evolve in multiple dimensions (e.g., temperature, precipitation, or both) and fluctuate in opposite directions. For example, a spring that is on average cooler than usual could have few very warm days. In addition, complex social, economic and political conditions drive weather perceptions4,5,8. However, it is the systematic one-sided biases whose in-depth understanding likely holds great potential for improved policy designs that are aimed at enhancing resilience in farm outcomes in response to climate uncertainty. Perception biases as motivated by land use decisions [Testing of Hypothesis 2] To further understand what underlies weather perception biases (as recorded in Table 4a–c) we investigate these in relation to farmers’ land use decisions in the Dakotas. Table 5a–c provides evidence (based on the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi^{2}$$\end{document}-test) for the dependence between perception biases and CRP land conversion decisions. The primary feature that underlies these results is a skew in the frequency of precipitation and drought beliefs when assessed in conjunction with farmers’ decision to cultivate lands that were initially enrolled under the CRP contract. To elaborate further, consider perception biases outlined in Table 5a–b.Table 5Contingency table between farmers’ decision to convert CRP lands for cropping (i.e., whether a farmer converted CRP land for cropping) and their beliefs about precipitation changes during 2004–2014.Land use decisionBeliefs about lower precipitation in 2014 relative to 2004Biased beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to H$$\end{document} )Consistent beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to L$$\end{document})Row Totals\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 (Yes)39(31)17(25)56\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 (No)83(91)84(76)167Column Totals122101223Test statistic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X\sim {\chi }_{1}^{2}$$\end{document})6.73(a)Land use decisionBeliefs about higher precipitation in 2014 relative to 2004Biased beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to L$$\end{document} )Consistent beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to H$$\end{document})Row Totals\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 (Yes)11(21)41(30)52\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 (No)81(71)96(106)177Column Totals92137229Test statistic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X\sim {\chi }_{1}^{2}$$\end{document})10.13(b)Land use decisionBeliefs about lower drought frequency in 2014 relative to 2004Biased beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to H$$\end{document} )Consistent beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to L$$\end{document})Row Totals\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 (Yes)10(20)60(50)70\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 (No)59(49)115(125)174Column Totals69175229Test statistic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X\sim {\chi }_{1}^{2}$$\end{document})9.48(c)Each cell in these tables provides ‘observed’ frequency and ‘expected frequency’ (see in the parentheses) for pairwise CRP land conversion decisions (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}) and weather change beliefs (bi).Expected frequency values are calculated under independence assumption between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} and bi. For cell (i,j) the expected frequency equals {column-total(i) x row-total(j)} ÷ (total sample size). For example, for cell (1,1) in Table 5a the expected frequency equals (122 × 56) ÷ (223) = 30.6.The critical value for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\chi }_{1}^{2}$$\end{document} at 95% confidence level is equal to 3.84. Table 5Contingency table between farmers’ decision to convert CRP lands for cropping (i.e., whether a farmer converted CRP land for cropping) and their beliefs about precipitation changes during 2004–2014.Land use decisionBeliefs about lower precipitation in 2014 relative to 2004Biased beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to H$$\end{document} )Consistent beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to L$$\end{document})Row Totals\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 (Yes)39(31)17(25)56\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 (No)83(91)84(76)167Column Totals122101223Test statistic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X\sim {\chi }_{1}^{2}$$\end{document})6.73(a)Land use decisionBeliefs about higher precipitation in 2014 relative to 2004Biased beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to L$$\end{document} )Consistent beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to H$$\end{document})Row Totals\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 (Yes)11(21)41(30)52\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 (No)81(71)96(106)177Column Totals92137229Test statistic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X\sim {\chi }_{1}^{2}$$\end{document})10.13(b)Land use decisionBeliefs about lower drought frequency in 2014 relative to 2004Biased beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to H$$\end{document} )Consistent beliefs(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L \to L$$\end{document})Row Totals\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 (Yes)10(20)60(50)70\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 (No)59(49)115(125)174Column Totals69175229Test statistic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X\sim {\chi }_{1}^{2}$$\end{document})9.48(c)Each cell in these tables provides ‘observed’ frequency and ‘expected frequency’ (see in the parentheses) for pairwise CRP land conversion decisions (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}) and weather change beliefs (bi).Expected frequency values are calculated under independence assumption between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} and bi. For cell (i,j) the expected frequency equals {column-total(i) x row-total(j)} ÷ (total sample size). For example, for cell (1,1) in Table 5a the expected frequency equals (122 × 56) ÷ (223) = 30.6.The critical value for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\chi }_{1}^{2}$$\end{document} at 95% confidence level is equal to 3.84. Contingency table between farmers’ decision to convert CRP lands for cropping (i.e., whether a farmer converted CRP land for cropping) and their beliefs about precipitation changes during 2004–2014. 39 (31) 17 (25) 83 (91) 84 (76) 11 (21) 41 (30) 81 (71) 96 (106) 10 (20) 60 (50) 59 (49) 115 (125) Each cell in these tables provides ‘observed’ frequency and ‘expected frequency’ (see in the parentheses) for pairwise CRP land conversion decisions (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}) and weather change beliefs (bi). Expected frequency values are calculated under independence assumption between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} and bi. For cell (i,j) the expected frequency equals {column-total(i) x row-total(j)} ÷ (total sample size). For example, for cell (1,1) in Table 5a the expected frequency equals (122 × 56) ÷ (223) = 30.6. The critical value for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\chi }_{1}^{2}$$\end{document} at 95% confidence level is equal to 3.84. More farmers misperceived that weather became drier (when the data shows it became wetter) as compared to those misperceiving that weather became wetter (when the station records show change in the opposite direction). Specifically, 122 out of 223 farmers perceived that weather became drier when it became wetter from 2004 to 2014 (Table 5a). In contrast, only 92 out of 229 farmers perceived was becoming wetter when it became drier during the same period (Table 5b). Assessing beliefs conditional on CRP land conversion decisions (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 or 0) reveals that the misperceptions are relatively concentrated in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1 category. In particlar, more than twice among converting farmers misperceived ‘drier’ weather (Table 5a), while only one-fourth among these misperceived ‘wetter’ conditions in 2014 (Table 5b). Farmers in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0 category, on the other hand, are relatively more uniformly spread out across consistent and biased perception categories (i.e., 84 versus 83 in lower precipitation case; and 81 versus 96 in higher precipitation case). A similar situation arises when we assess drought perceptions conditional on CRP land conversion decisions, see Table 5c. Perception biases that ‘precipitation was higher’ or ‘droughts were fewer’ (relative to data-based records) were predominant among farmers who decided to convert CRP lands for cultivation. Lands that are enrolled under CRP contracts are typically of medium or low productivity lands51. Hypothesis II provides a conjecture that once a farmer opts out of the CRP contract she might be motivated to believe in favorable weather conditions relative to what the data supports, and such (mis)perceptions might be greater when soils are less conducive for cropping28. We find some evidence in support for Hypothesis II by rejecting statistical independence between weather beliefs and land use decisions at 95% confidence. However, this does not guarantee that past decisions caused perception biases becasue of potential endogeneity issues explained earlier. Below we outline the results of our IV-based analysis of the impact of land use decisions on weather change perceptions. Motivated beliefs: land use decisions as a potential driver of weather perception biases Tables 6, 7 present results from our 2SLS framework (outlined in Fig. 3 and discussed in section "Assessing potential causal impact channel between land use decisions and weather perception biases") to establish that Dakotas’ farmers are likely to misperceive weather changes after converting their CRP lands for cultivation. In Table 6 (first-stage regression) we show that one-dollar increase in per-acre CRP rents would reduce the odds of converting CRP lands by 4% (significant at α = 0.01). On the other hand, 1% higher ratio of cropland-to-grassland return increases the odds of CRP land conversion by 8%, however the latter impact is statistically insignificant. Further, CRP lands were relatively more likely to be highly erodible lands, which is in line with the policy mandate of CRP (discussed earlier), and among older farmers in higher income categories.Table 6Binomial logit regression for farmers’ CRP land conversion decision (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document}).VariableDependent Variable: ln[Pr(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} = 1)/ Pr(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} = 0)]rent− 0.04***(0.01)ratio0.08(0.17)hel0.02***(0.005)age− 0.21***(0.08)educ0.13(0.09)inc0.43***(0.07)Intercept− 2.00***(0.74)-2LogL934Response counts\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1: 226; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0: 714N940*p < 0.1, **p < 0.05, ***p < 0.01.Table 7Binomial logit regression for farmers’ precipitation and drought beliefs (i.e., bi).VariablePrecipitation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \le - 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{L} \in \{ L \to H,L \to L\}$$\end{document}Precipitation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \ge 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{H} \in \{ H \to L,$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to H\}$$\end{document}Drought:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document} < 0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{L} \in \{ L \to H,L \to L\}b_{i}^{L} \in \{ L \to H,L \to L\}$$\end{document}IVnon-IVIVnon-IVIVnon-IV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{p}_{i}^{(CRP \to crop)}$$\end{document}0.39***(0.15)− 0.41**(0.20)− 0.60***(0.18)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}0.82**(0.36)− 0.91**(0.39)− 0.88***(0.31)age0.49***(0.16)0.36**(0.14)− 0.36**(0.16)− 0.21(0.14)− 0.52***(0.14)− 0.29**(0.12)educ0.29*(0.18)0.40**(0.17)− 0.05(0.17)− 0.14(0.15)0.03(0.15)− 0.15(0.14)inc− 0.24***(0.15)− 0.04(0.12)− 0.03(0.18)− 0.25**(0.11)− 0.12(0.16)− 0.26***(0.10)Intercept− 2.38**(0.94)− 2.21**(0.93)− 3.58***(0.63)1.80**(0.86)− 1.95***(0.74)1.90**(0.75)-2LogL288291295294411415Total N214223333Biased#Accurate#1189011596133218*p < 0.1, **p < 0.05, ***p < 0.01. Standard errors in parentheses. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document} > 0 case was excluded due to low count of locations (only 76) where droughts intensified during 2004–2014. Finally, temperature perceptions were not found to be associated with CRP land conversion decisions in a statistically significant manner. Table 6Binomial logit regression for farmers’ CRP land conversion decision (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document}).VariableDependent Variable: ln[Pr(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} = 1)/ Pr(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document} = 0)]rent− 0.04***(0.01)ratio0.08(0.17)hel0.02***(0.005)age− 0.21***(0.08)educ0.13(0.09)inc0.43***(0.07)Intercept− 2.00***(0.74)-2LogL934Response counts\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 1: 226; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document} = 0: 714N940*p < 0.1, **p < 0.05, ***p < 0.01. Binomial logit regression for farmers’ CRP land conversion decision (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l$$\end{document}). − 0.04*** (0.01) 0.08 (0.17) 0.02*** (0.005) − 0.21*** (0.08) 0.13 (0.09) 0.43*** (0.07) − 2.00*** (0.74) *p < 0.1, **p < 0.05, ***p < 0.01. Table 7Binomial logit regression for farmers’ precipitation and drought beliefs (i.e., bi).VariablePrecipitation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \le - 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{L} \in \{ L \to H,L \to L\}$$\end{document}Precipitation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14} \ge 5\%$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{H} \in \{ H \to L,$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H \to H\}$$\end{document}Drought:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document} < 0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{i}^{L} \in \{ L \to H,L \to L\}b_{i}^{L} \in \{ L \to H,L \to L\}$$\end{document}IVnon-IVIVnon-IVIVnon-IV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{p}_{i}^{(CRP \to crop)}$$\end{document}0.39***(0.15)− 0.41**(0.20)− 0.60***(0.18)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}0.82**(0.36)− 0.91**(0.39)− 0.88***(0.31)age0.49***(0.16)0.36**(0.14)− 0.36**(0.16)− 0.21(0.14)− 0.52***(0.14)− 0.29**(0.12)educ0.29*(0.18)0.40**(0.17)− 0.05(0.17)− 0.14(0.15)0.03(0.15)− 0.15(0.14)inc− 0.24***(0.15)− 0.04(0.12)− 0.03(0.18)− 0.25**(0.11)− 0.12(0.16)− 0.26***(0.10)Intercept− 2.38**(0.94)− 2.21**(0.93)− 3.58***(0.63)1.80**(0.86)− 1.95***(0.74)1.90**(0.75)-2LogL288291295294411415Total N214223333Biased#Accurate#1189011596133218*p < 0.1, **p < 0.05, ***p < 0.01. Standard errors in parentheses. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document} > 0 case was excluded due to low count of locations (only 76) where droughts intensified during 2004–2014. Finally, temperature perceptions were not found to be associated with CRP land conversion decisions in a statistically significant manner. Binomial logit regression for farmers’ precipitation and drought beliefs (i.e., bi). 0.39*** (0.15) − 0.41** (0.20) − 0.60*** (0.18) 0.82** (0.36) − 0.91** (0.39) − 0.88*** (0.31) 0.49*** (0.16) 0.36** (0.14) − 0.36** (0.16) − 0.21 (0.14) − 0.52*** (0.14) − 0.29** (0.12) 0.29* (0.18) 0.40** (0.17) − 0.05 (0.17) − 0.14 (0.15) 0.03 (0.15) − 0.15 (0.14) − 0.24*** (0.15) − 0.04 (0.12) − 0.03 (0.18) − 0.25** (0.11) − 0.12 (0.16) − 0.26*** (0.10) − 2.38** (0.94) − 2.21** (0.93) − 3.58*** (0.63) 1.80** (0.86) − 1.95*** (0.74) 1.90** (0.75) Biased# Accurate# *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors in parentheses. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document} > 0 case was excluded due to low count of locations (only 76) where droughts intensified during 2004–2014. Finally, temperature perceptions were not found to be associated with CRP land conversion decisions in a statistically significant manner. Table 7 (second-stage regression) presents our results on the potential causal impact of the likelihood of CRP land conversion (as estimated from first-stage regression) on the odds of biased weather beliefs relative to the odds of consistent beliefs. We find that biases in precipitation and drought perceptions were statistically significantly affected by the decision to cultivate crops on CRP lands. Specifically, a higher likelihood of CRP land conversions led to an average of 39% higher odds (significant at α = 0.01) of misperceiving drier conditions; 41% (α = 0.01) lower odds of misperceiving wetter conditions; and 60% (α = 0.01) lower odds of misperceiving less droughts during 2004–2014 (Table 7). Note that the impact estimate for land conversion decision, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta l_{i}$$\end{document}, was lower (in absolute value) for IV-based regression relative to the non-IV case, which reveals a correction of omitted variable bias in the non-IV case due to endogeneity between farmers’ decisions and weather perception biases. Issues in identifying systematic biases in weather change perceptions Weather change misperceptions are measured as the difference between farmer perceptions and actual weather changes in their locality. Therefore, whether, and to what extent, farmers are found to have misperceptions about local weather depends on the accuracy of survey-based perception reports and past weather records. Also, our causal inference analysis relies on the instruments used in Eq. (3) and so our analytical results will be affected by issues related to the validity or strengths of the IVs used. Below we discuss the details related to the impacts of these potential issues and outline our strategies to check for robustness of our findings. Robustness checks related to measurement of perceptions and weather records We assess the potential issues related to the measurement of weather change misperceptions: (i) possibly inaccurate farmer recall of past weather changes, (ii) alternative sources of local weather records, and (iii) a threshold-based definition of weather changes for comparing discrete perception categories with continuous weather records (please see section "Articulating biases in past weather change perceptions"). A farmer’s perception of long-term weather trends (e.g., over a decade) can be influenced by short-term weather fluctuations or recent salient events such as droughts33,52. As a check for robustness, we account for limited recall capacity for past weather events through the use of two auxiliary measures of weather conditions at the beginning of our study period. First, we assign 2004 weather to be an average of weather records in 2003, 2004 and 2005 to evaluate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w$$\end{document} in model (1). Second, we evaluate how recent salient events, i.e., 2006 drought and heat waves53–55 and 2011 floods56, 57, might influence farmers’ weather perceptions58,59. To do so, we calculate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta w$$\end{document} based on the initial weather conditions in 2006 and 2011, in addition to those in 2004. Tables S1–S3, SM, show that, unlike \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta p^{04 - 14}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta d^{04 - 14}$$\end{document}, precipitation and drought perceptions were not significantly correlated with past weather changes during 2006–2014 and 2011–2014 periods, respectively, in a statistically manner. Temperature perceptions were not significantly correlated with data-based evidence for any of the initial years, i.e., 2004–2014 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{04 - 14}$$\end{document}), 2006–2014 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{06 - 14}$$\end{document}), or 2011–2014 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t^{11 - 14}$$\end{document}). Second, we utilize an alternative and independent source of local weather records, i.e., PRISM weather records made available by Oregon State University to estimate model (1) and evaluate biases in farmer perceptions. Specifically, we utilize the PRISM data (Oregon State University), which is organized in 4 km × 4 km grids, besides the GHCN weather station records. We find coefficient estimates for model (1) to be very similar irrespective of the source of weather data (except that model (1) exhibits better fit with station-level data). Relevant details are provided in SM, section S2. We present our main results based on weather station data because people (including farmers) commonly learn about local weather from meteorological stations. Finally, we assess how weather misperceptions are influenced by alternate threshold values, i.e., ± 5%, ± 10% and ± 20%, which are used to define little or no change in local weather records during 2004–2014. As discussed earlier, these thresholds are necessary to convert continuous weather indicators into categorical variables so that they are comparable with the survey-based perception categories. Obviously, the threshold values that we set will affect perception bias measurements. We examine the robustness of our analysis to distinct threshold values in two ways. We first assess our threshold values relative to data-based range in precipitation and temperature changes during 2004–2014. For this purpose, we compare May–July weather anomalies in 2004 and 2014 against a common average weather baseline during 1991–2020, made available by PRISM web-portal. We find total precipitation and average temperatures in 2014 to be on-average lower than 2004 levels. Specifically, the spatial range of precipitation changes across counties was − 20% to − 30%, and the spatial range of temperature changes across counties was − 2.5% to − 12.5%. Please see SM, section S3 for details. Therefore, sensitivity analyses using 5%–20% as the threshold values for ‘little or no change’ category provide an appropriate spatial coverage of recorded weather changes in our study region. Second, we conduct a sensitivity analysis and find that distribution of weather misperceptions was robust across alternate thresholds. For example, the distribution of respondents who misperceived greater precipitation as “drier” (i.e., 22% farmers), and those who misperceived less precipitation as “wetter” (i.e., 34% farmers) remain the same across the three thresholds. Moreover, the linkage between past land use decisions and perception biases also remains consistent in the 2SLS framework irrespective of threshold choice. Please see SM, section S3 for details. Specification tests for overidentifying restrictions and weak instruments We conduct specification tests for overidentifying restrictions and weak instruments with reference to our 2SLS framework in Eqs. (3)-(5). The Sargan test for overidentifying restrictions was implemented as described in 27, pp. 123. The concern for overidentification arises due to the fact that we include two instruments for one endogenous variable (i.e., one additional instrument than necessary) in the first stage regression Eq. (3). Null hypothesis of the Sargan test is that the instruments are correctly excluded and that these are uncorrelated with the error terms in second-stage regression models for perception biases (i.e., Eqs. (4)-(5) against the alternate hypothesis that instruments are invalid. Further, Staiger and Stock60 provided a decision rule for weak instruments such that when the regression F-statistic is less than 10 we infer that the instrument is weak. Overall, we concluded that our instrumetns, namely renti and ratioi, are correctly excluded (with 95% confidence) from the second stage regression models (4)-(5) and that these instruments are not weak for estimating the 2SLS regression system (3)-(5). Please see the Results section and SM, section S6 and Table S7 for further details. Finally, as robustness check, we consider a different instrument, namely difference in county-level soil-specific rental rates (SRR) between 2004 and 2014, that is, we use srri in place of renti. The SRRs are based on county-average non-irrigated cropland rental rates and are further adjusted to parcel-level soil productivity61,62. The SRR of a CRP application is used to set the maximum annual payment rate of the application, thereby providing a program-based payment rate. Most CRP applications have a rental rate around the average SRR61, and we find high correlation between srri and renti (correlation coefficient = 0.76; p < 0.01). Among the CRP signups in our data, the average parcel-level srri pre-2007 period was 46$/acre and then rose sharply to 77 $/acre post-2007 (i.e. 2010 and 2011). Details of 2SLS estimation results with this distinct IV is presented in SM, Table S12. Overall, both stages of our 2SLS framework is robust to the use of alternative instruments, as particularly reflected through the distribution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{p}_{i}^{(CRP \to crop)}$$\end{document} that remains similar irrespective of the choice of IVs. Discussions and conclusion There is a well-established literature that posits perceptions of climate change and weather patterns as critical inputs for designing effective policies for climate change mitigation and adaptation. A fundamental impediment in advancing the objectives of climate policy is that individual beliefs may not be consistent with scientific evidence63,64. Recent studies of climate change beliefs highlight the interactions between spatial climate patterns and heterogeneous individual interests/values that are tied to local resource constraints and socio-cultural factors (e.g.,24,65). We extend this line of research by analysing beliefs that may be distinct across weather indicators such as temperature, precipitation and droughts. The basis of our inquiry is the complex composition of climate or weather information such that individuals may attribute climate change to spatio-temporal changes in temperature, precipitation, or both, depending on the local or regional contexts. We study farm-level heterogeneity in perception biases for three weather indicators: temperature, precipitation, and drought frequency, in relation to local meteorological evidence and farmers’ socioeconomic and geographic information. Our analysis highlights that farmers may perceive weather indicators distinctly from each other. In fact, farmers in our sample perceived local weather changes in ways that reflects on the impact of each weather indicator on their production system. We also provide evidence on the potential causal impact channel between farmers’ land use decisions and weather beliefs. Having made the decision to pursue crop production on protected lands seemed to motivate farmers to systematically ignore adverse weather signals of the past. In the existing literature on motivated climate change reasoning, prior beliefs and political polarization were documented as drivers of climate change beliefs. Our result that links weather perceptions to past land use decisions adds a new economic perspective to the understanding of climate change beliefs and relevant communication strategies (as originally proposed by66). These authors suggested that identifying individual’s goals or motivation for belief distortion can facilitate effective climate change communication. Our findings suggest an indirect economic motivation (as mediated through land use decisions) for climate change beliefs that may present a challenge for communicating scientific evidence when high commodity prices favor crop-based production. We point to a couple of caveats in our findings. Our analysis is based on a cross-sectional sample of farmers’ weather perceptions in a specific region. While the production setting in our study region is quite homogeneous and the sample is relatively large and representative, our cross-sectional models for weather perceptions are susceptible to non-weather confounders that may influence how farmers’ beliefs shift over time. It is also important to study weather perception biases in other regions and contexts (e.g., alternative land use types or farm management decisions). More importantly, climate and weather perceptions are based on complex cognitive processes and so in-depth examination is necessary with methods and concepts in psychology and cognitive sciences to pinpoint key factors and mechanisms that drive perceptions. Regardless, our findings suggest that climate policy designs that solely rely on objective climate and weather data might be misguided. It will be beneficial for policymakers to devise targeted climate mitigation and adaptation strategies that take into account farmers’ perceptions and beliefs. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1