How the sausage is made: Testing the effectiveness of an informative video in promoting sustainable food consumption Authors: Lisa A Kramer, Peter Landry Published: 2026 ABSTRACT AbstractWe use an incentivized consumption measure to assess the impacts of watching a roughly 15-minute video about industrial-scale pork production on sustainable food consumption. We find that the demand for (more sustainable) plant-based meals increases by 252.7 percent while the demand for meat-containing meals decreases by 27.0 percent immediately after watching the video. Notably, these effects exhibit some persistence. Compared to baseline levels, we find that demand at a one-week follow-up is still 121.8 percent higher for plant-based meals and 13.5 percent lower for meat-containing meals. The video appears to have a particularly large impact on the demand for pork, with immediate and one-week decreases of 45.2 and 30.1 percent. We also observe smaller yet still significant decreases in the demand for other meat products, with immediate and one-week reductions of 24.0 and 10.6 percent, suggesting that information about industrial-scale pork production can affect preferences for other types of meat. Overall, the video appears to be quite effective in promoting plant-based eating compared to others tested in the literature. Even so, the relatively low baseline popularity of pork compared to other meats in our sample suggests that the impact of the video may be limited by its focus on pork, raising the possibility that a broader portrayal of industrial meat production could be even more effective in promoting sustainable food consumption.Subject terms: Human behaviour, Psychology and behaviour Introduction It is well known that eating plant-based foods instead of animal-based foods can carry substantial environmental benefits. Animal agriculture is a leading contributor to climate change, accounting for roughly 15 percent of global greenhouse gas emissions1,2 For comparison, the transportation sector (including cars, buses, trucks, and planes) has also been estimated to account for roughly 15 percent of global greenhouse gas emissions3. Animal agriculture is also a primary driver of habitat destruction, including deforestation in the Amazon, and depletion of natural resources4–7. Plant-based eating can provide other benefits too. For instance, high rates of meat consumption, particularly of red meat, have been linked to many human health ailments8–10. Naturally, a shift towards plant-based eating can also alleviate the large-scale harms to animal welfare from the industrial-scale production of meat and other animal products11–13. In light of these impacts, researchers across many disciplines have become interested in identifying effective ways of promoting plant-based eating. One approach considered in this literature is the use of videos. Informal surveys of consumers on plant-based diets allude to the possibility that videos could be effective in promoting plant-based eating (in survey results described by Orzechowski14 and Polanco15, for instance, videos were the most commonly cited impetus for adopting a plant-based diet). However, controlled experimental tests tell a different story. Notably, the large majority of videos tested in this literature have had no observable effect on dietary choices16–22,23, while any observed effects appear to be short-lived24,25,26. There are also studies that test the effects of videos on other measures besides consumption (e.g., on attitudes, intentions, or donations, as in Mercy For Animals27, Pabian et al.28, Espinosa et al.29, and Thomas et al.30) and studies in which watching a video is an endogenous outcome measure31,32). How can we reconcile the apparent ineffectiveness of videos in promoting plant-based eating, as revealed in experimental tests, with survey evidence suggesting that videos may be a main impetus for dietary change? One possible explanation for the discrepancy is that perhaps some videos can meaningfully impact viewers’ dietary choices, but these videos are outliers relative to the vast majority that have very little effect. If true, it would be useful to identify these outliers. One candidate is Dominion, a 2018 Australian documentary that provides an in-depth look at the use of animals in agriculture and other commercial applications. According to IMDb.com, Dominion has earned 15 awards, including Best Picture at the 2018 Los Angeles Independent Film Festival, and its 8.9/10 average viewer rating puts it in the 99.85th percentile of all full-length movies with at least 1,000 user ratings as of March 19, 2025. At that time, Dominion had the 66th highest rating (and 6th highest among documentaries) out of all 45,196 full-length movies with at least 1,000 user ratings on IMDb.com. In this study, we conduct a randomized controlled experiment to test the effectiveness of a 16-minute segment from the beginning of Dominion in promoting plant-based eating. This segment focuses on the industrial-scale production of pigs, while later portions of the film, addressing chickens, cows, and other animals, are not included. The segment we use does not address the environmental impacts of industrial-scale pork production. Instead, it describes the experiences of the pigs in these systems, and in doing so, provides information concerning the impacts on the pigs’ welfare. While it may seem unnatural to promote sustainable food consumption without appealing to sustainability, previous work suggests that appeals based on environmental impacts may not be the most effective messaging for promoting plant-based eating33,34. Our experiment had two parts. The first part was an in-person session in which participants completed a survey with their assigned video—either the treatment video segment from Dominion or a control video addressing the role of mushrooms in forest ecosystems—embedded in the survey. The second part was an online survey that was emailed to each participant exactly seven days after the participant completed the in-person session. To measure the impacts of the treatment video, we used an incentivized measure of consumption in which participants chose a menu item from a local café at three different times: before watching their assigned video, immediately after watching the video, and in the part 2 follow-up survey sent one week later. Each of these choices involved selecting a base meal and then selecting a protein to accompany the base. Of particular importance for our study, participants faced the same five protein options regardless of their chosen base: bacon, chicken, steak, tofu, and none. Finally, these choices were incentivized in that participants were aware that at the end of the experiment roughly 50 percent of them would win a voucher for a designated menu item corresponding to one of their selections in the experiment. Using this measure, we find that watching the treatment video leads to statistically significant increases in the demand for meatless meals, by 252.7 percent immediately after watching the video and by 121.8 percent at the one-week follow-up, with immediate and one-week decreases in the demand for meat-containing meals of 27.0 percent and 13.5 percent, respectively. These effects after one week are particularly noteworthy considering the lack of prior experimental evidence indicating that watching a video can affect meat consumption one week later. Results Participant characteristics Study participants were recruited from the Rotman School of Management’s credit pool, comprised of undergraduate students in introductory commerce courses who participate in research for course credit. We provide a summary of relevant participant characteristics in Table 1. Naturally, our student sample was considerably younger than the general population. While female participants outnumbered male participants, females were slightly under-represented (and males slightly over-represented), relative to national averages for post-secondary students in Canada35. While not shown here, our sample was also diverse in terms of race/ethnicity and religion/creed (see Appendix Tables 7 and 8 for more details). Table 1Participant characteristics.TreatmentControlAll# Participants5695801149% Attrition to part 22.83.83.3mean days to part 2 (SD)7.69 (1.57)7.55 (1.22)7.62 (1.41)% Female59.455.957.6% Male40.644.142.4mean age (SD)19.38 (1.33)19.43 (1.44)19.41 (1.38)% Win voucher53.650.552.0% Collect voucher (if win)53.451.953.3This table contains summary statistics on participant characteristics, by treatment group, by control group, and in total, beginning with the number of participants. “% attrition to part 2” is the percentage of participants who participated in part 1 but not part 2 of the study. “Mean days to part 2 (SD)” is the average number of days (and standard deviation in the number of days) that lapsed between dates when a participant completed parts 1 and 2 of the study. “% female” and “% male” reflect the percentage of participants who reported their sex as female or male. Participants had the option to instead select “other sex” but none did. “Mean age (SD)” captures the average (and standard deviation) of the age of participants in years. “% win voucher” is the percentage of participants who were allocated their selected food choice in the incentivized consumption measure. This differs from the targeted 50% success rate due to random chance. Finally, “% collect voucher (if win)” is the percentage of participants who won a voucher and returned to the lab to pick up the voucher.The percentages of collected vouchers in the treatment and control groups shown here reflect lower bounds on the actual percentages. This is because there were seven instances in which we were unable to identify whether a voucher was collected by a treatment group participant or a control group participant (this could happen for various reasons, such as the on-duty lab research assistant forgetting to record the voucher number picked up by the student or recording an invalid number). Note that these additional instances are still included in the percentage of collected vouchers among all participants. Participant characteristics. This table contains summary statistics on participant characteristics, by treatment group, by control group, and in total, beginning with the number of participants. “% attrition to part 2” is the percentage of participants who participated in part 1 but not part 2 of the study. “Mean days to part 2 (SD)” is the average number of days (and standard deviation in the number of days) that lapsed between dates when a participant completed parts 1 and 2 of the study. “% female” and “% male” reflect the percentage of participants who reported their sex as female or male. Participants had the option to instead select “other sex” but none did. “Mean age (SD)” captures the average (and standard deviation) of the age of participants in years. “% win voucher” is the percentage of participants who were allocated their selected food choice in the incentivized consumption measure. This differs from the targeted 50% success rate due to random chance. Finally, “% collect voucher (if win)” is the percentage of participants who won a voucher and returned to the lab to pick up the voucher. The percentages of collected vouchers in the treatment and control groups shown here reflect lower bounds on the actual percentages. This is because there were seven instances in which we were unable to identify whether a voucher was collected by a treatment group participant or a control group participant (this could happen for various reasons, such as the on-duty lab research assistant forgetting to record the voucher number picked up by the student or recording an invalid number). Note that these additional instances are still included in the percentage of collected vouchers among all participants. Table 7Participant race/ethnicity. % reporting inTreatmentControl AllArab2.82.92.9Black2.61.42Indigenous0.40.20.3Latin American1.81.71.7Other23.426.324.8South Asian24.323.523.9Southeast Asian28.527.828.2West Asian31.92.4White19.422.120.7 None Reported 0.2 0.2 0.2 This table shows the percent of participants who selected each option for the race/ethnicity question in part 1. The sum of the percentages in a given column may exceed 100 because some participants selected more than one option. Participant race/ethnicity. This table shows the percent of participants who selected each option for the race/ethnicity question in part 1. The sum of the percentages in a given column may exceed 100 because some participants selected more than one option. Table 8Participant religion/creed. % reporting inTreatmentControl AllAgnostic11.110.710.9Atheist16.715.316Buddhist6.28.47.3Christian23.924.524.2Hindu7.78.38Indigenous spirituality0.40.20.3Jewish0.91.91.4Muslim9.77.98.8Other18.120.219.1Sikh2.11.21.7 None reported 3.3 1.4 2.3 This table shows the percent of participants who selected each option for the religion/creed question in part 1. Participant religion/creed. This table shows the percent of participants who selected each option for the religion/creed question in part 1. As we can see in Table 1, attrition from the in-person session to the part 2 follow-up survey was low. Participants were generally prompt in completing the part 2 follow-up survey (on average, completing the follow-up survey within a day of receiving the link that was sent seven days after completing part 1). Meanwhile, around half of all participants won a voucher for a meal that they selected in one of their three incentivized consumption choices. Of these, a little more than half returned to the lab to collect their voucher. In total, 319 vouchers were distributed, with over 95 percent of these redeemed for free meals (based on validated invoices from the café we partnered with). Main results: effects of videos on incentivized consumption Table 2 summarizes the immediate effects of the videos on our incentivized consumption measure. As seen, a significant 24.4 percentage-point reduction in the demand for meat-containing meals was observed in the treatment group, with no significant change in the control group (moreover, as shown in the bottom panel of Table 2, the difference between the treatment and control group effects is highly significant). When we separate the overall treatment group effect into its effects on the demand for pork and on the demand for other types of meat, we see a smaller 5.8 percentage-point reduction in the demand for pork compared to the 18.6 percentage-point reduction in the demand for other meat among treatment group participants. However, this discrepancy is largely a reflection of the fact that participants were much less likely to choose a meal with pork (bacon) than to choose a meal with beef or chicken in their initial consumption choice. In terms of the proportional change, there was a larger, 45.2 percent reduction in demand for pork among treatment group participants compared to a 24.0 percent reduction in demand for other types of meat; meanwhile, there was a 27.0 percent proportional reduction in demand for all meat-containing meals, while demand for meatless meals increased by 252.7 percent (from 9.7 to 34.1 percent of all post-video voucher choices in the treatment group). Table 2Immediate effects on protein choice.GroupProtein% of voucher choicesp-valuePre-videoPost-videoChange Treatment ()Pork12.87.0–5.8Other meat77.558.9–18.6Total meat90.365.9–24.4Meatless9.734.124.4 Control()Pork11.913.11.20.071Other meat79.577.8–1.70.033Total meat91.490.9–0.50.366Meatless8.69.10.50.366Difference(T−C)pork0.9-6.1–7.0Other meat–2.0 –18.9–16.9Total meat–1.0 –25.0 –23.9 Meatless1.025.023.9The top two panels in this table show the respective within-group effects of the treatment and control videos on protein choices immediately after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the immediate treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Immediate effects on protein choice. Control () Difference (T−C) The top two panels in this table show the respective within-group effects of the treatment and control videos on protein choices immediately after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the immediate treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Table 3 shows the effects of the videos on our primary consumption measure after one week. Here, we see a significant decrease in the demand for meat-containing meals among treatment group participants one week after watching the treatment video. This overall reduction in the demand for meat reflects significant reductions in the demand for pork as well as in the demand for other types meat. Meanwhile, no such effects are observed in the control group, while the difference between the treatment and control group effects remains highly significant. With that said, we can see a bit of decay in the size of the treatment group effects; for instance, the 12.1 percentage-point reduction in demand for all meat among treatment group participants after one week is roughly half as large as the immediate 24.4 percentage-point reduction in demand seen in Table 2. Expressed as proportions, we observe a 13.5 percent decrease in demand for all meat-containing meals—including a 30.1 percent decrease for pork and a 10.6 percent decrease for other types of meat—among treatment group participants after one week, offset by a 121.8 percent increase in demand for meatless meals. Table 3Effects on protein choice at one-week follow-up.GroupProtein% of voucher choicesp-valuePre-videoFollow-upChange Treatment()Pork13.29.20.004Other meat76.968.7Total meat90.177.9Meatless9.922.112.1 Control()Pork12.013.10.288Other meat79.277.40.141Total meat91.290.50.285Meatless8.89.50.70.285 Difference(T−C)Pork1.2-3.90.003Other meat–2.4 –8.7 0.004Total meat–1.2 –12.6 Meatless1.212.611.4The top two panels of this table show the respective within-group effects of the treatment and control videos on protein choices one week after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the one-week treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Note that sample sizes are less than those presented in Table 2 due to attrition from part 1 to part 2. Effects on protein choice at one-week follow-up. Treatment () Control () Difference (T−C) The top two panels of this table show the respective within-group effects of the treatment and control videos on protein choices one week after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the one-week treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Note that sample sizes are less than those presented in Table 2 due to attrition from part 1 to part 2. The observed effects of the treatment video on overall meat consumption in Tables 2 and 3 do not appear to be driven by shifts in preferences for particular base meal/protein combinations. For all four base meals, there were similar propensities to choose a meat as the protein option (81.6% for arugula salad, 84.3% for asado bowl, 85.5% for bagel sandwich, and 81.6% for couscous salad, across all consumption choices in both groups). Furthermore, compared to the baseline consumption choices, there was a reduced propensity to choose meat both immediately and one week after watching the treatment video for each of the four base meal options. For more detailed breakdowns of participants’ base meal selections and associated protein selections, see Appendix Tables 9, 10 and 11. Table 9Immediate effects on base meal. GroupMeal% of voucher choicesp-valuePre-videoPost-videoChangeTreatment()Arugula Asado Bagel Couscous Control()Arugula Asado Bagel Couscous Difference(T−C)Arugula Asado Bagel Couscous The top two panels in this table show the respective within-group effects of the treatment and control videos on base meal choices immediately after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the one-week treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Immediate effects on base meal. Treatment () Control () Difference (T−C) The top two panels in this table show the respective within-group effects of the treatment and control videos on base meal choices immediately after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the one-week treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Table 10One-week effects on base meal. GroupMeal% of voucher choicesp-valuePre-videoPost-videoChangeTreatment()Arugula Asado Bagel Couscous Control()Arugula Asado Bagel Couscous Difference(T−C)Arugula Asado Bagel Couscous The top two panels in this table show the respective within-group effects of the treatment and control videos on base meal choices one week after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the one-week treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Note that sample sizes are less than those presented in Table 9 due to attrition from part 1 to part 2. One-week effects on base meal. Treatment () Control () Difference (T−C) The top two panels in this table show the respective within-group effects of the treatment and control videos on base meal choices one week after watching the assigned video (relative to participants’ baseline choices). For these (before and after) comparisons, two-sided paired t-tests were used to calculate p-values. The bottom panel shows the differences between the one-week treatment group effects and the corresponding control group effects. The p-values associated with these difference-in-difference estimates were calculated using independent sample t-tests (also two-sided). Note that sample sizes are less than those presented in Table 9 due to attrition from part 1 to part 2. Table 11Base meal/protein associations. % of voucher choicesBase mealTofuNoneBaconBeefChickenTotal meatArugula salad Asado bowl Bagel sandwich Curried couscous Percentages reported in this table were calculated using all voucher choices (baseline, immediate post-video, one-week post-video) among all participants (treatment and control) who participated in both parts 1 and 2. Base meal/protein associations. Percentages reported in this table were calculated using all voucher choices (baseline, immediate post-video, one-week post-video) among all participants (treatment and control) who participated in both parts 1 and 2. A closer look at treatment effects To get a better understanding of what might drive changes in consumption among treatment group participants, we regressed several participant-level measures on an indicator equal to 1 if the participant selected a meal without meat immediately after watching the treatment video and 0 if the participant selected a meal with meat. In this regression, the subsample included only treatment group participants who chose a meal with meat in their initial (pre-video) consumption choice. The dependent variable can therefore be understood as an indicator of whether the participant switched from a meat option to a meatless option after watching the treatment video. Meanwhile, the participant-level regressors in this regression include: an indicator of whether the participant is female; an indicator of whether the participant chose a meal with bacon in the initial (pre-video) consumption choice; a measure of the participant’s “mindset flexibility”; an indicator of whether the participant opted to stop watching the video during a pause (after 11 min) in which the participant was given an explicit choice of whether to watch the last five minutes; a measure of the participant’s comprehension of content from their assigned video; measures of various emotional responses to the video; a measure capturing how much the participant learned from the video; a measure capturing the degree to which the participant is glad to have watched the video; a measure capturing the level of attention to the video; and a measure capturing the extent to which the video changed the participant’s worldview (more detailed explanations of these measures are provided in Methods). We present the regression estimates in Table 4. As seen, four of the regressors have a significant, positive effect in predicting switches to a meatless meal after watching the treatment video. First, choosing a meal with bacon (as opposed to chicken or beef) in the initial consumption choice is associated with a higher likelihood of switching. This is consistent with our previous finding that the treatment video has a proportionally larger effect on the demand for pork compared to its effect on the demand for other types of meat. Second, being female is associated with a higher likelihood of switching to a meatless meal. This finding is in keeping with prior evidence that women tend to be more open to plant-based eating than men36–39. Third, a higher degree of self-reported worldview change is associated with switching to a meatless meal. While it may seem intuitive that changes in worldview and changes in consumption would be linked, other arguably intuitive predictors of dietary change, such as higher (self-reported) levels of learning and attention, did not have a significant effect. Lastly, feeling ashamed after watching the treatment video predicts switches to a meatless meal. Note, however, that none of the other emotions tested had a significant effect. Table 4Dietary change regression results.CoefficientStd. errorp-valueConstantFemaleBacon in 1st choiceMindset flexibilityStop watchingComprehensionSadAngryBoredDisgustedSurprisedAshamedAmount learnedGlad watchedAttention to videoWorldview change—part 1In this table, the sample consists of 514 treatment-group participants who chose a meat item in their initial voucher choice. The dependent variable in the regression is an indicator equal to 1 if the participant switched to a vegetarian item in their (immediate) post-video voucher choice and a 0 otherwise. Dietary change regression results. In this table, the sample consists of 514 treatment-group participants who chose a meat item in their initial voucher choice. The dependent variable in the regression is an indicator equal to 1 if the participant switched to a vegetarian item in their (immediate) post-video voucher choice and a 0 otherwise. As a robustness check, we considered a variation of this regression in which all regressors were converted to binary measures based on a median split. That is, responses to questions with more than two options were converted to binary measures based on whether a participant’s response was above or below the median response for that measure; responses equal to the median were either included with below-median responses or above-median responses depending on which of these classifications resulted in a more equal distribution (for that measure). The results of this regression are reported in Appendix Table 12. In this version of the regression with binarized regressors, we still find that feeling ashamed, worldview change, being female, and choosing a meal with bacon in the initial consumption choice all have significant positive effects on the propensity to switch to a meatless meal after watching the treatment video. Meanwhile, all of the other regressors still have no effect. Table 12Dietary change regression results with binarized regressors. CoefficientStd. errorp-valueConstant Female Bacon in 1st choice Mindset flexibility Stop watching Comprehension Sad Angry Bored Disgusted Surprised Ashamed Amount learned Glad watched Attention to video Worldview change—part 1 For this table, the sample consists of 514 treatment group participants who chose a meat item in their initial voucher choice. The dependent variable in the regression is an indicator equal to 1 if the participant switched to a meatless item in their (immediate) post-video voucher choice and a 0 otherwise. As described in the text, the non-binary regressors in the regression reported in Table 4 were converted to binary measures in this regression, based on whether the response was above or below the median. Dietary change regression results with binarized regressors. For this table, the sample consists of 514 treatment group participants who chose a meat item in their initial voucher choice. The dependent variable in the regression is an indicator equal to 1 if the participant switched to a meatless item in their (immediate) post-video voucher choice and a 0 otherwise. As described in the text, the non-binary regressors in the regression reported in Table 4 were converted to binary measures in this regression, based on whether the response was above or below the median. Secondary analyses Table 5 summarizes the effects of the videos on various other participant-level measures besides consumption. To facilitate comparisons, in this table all measures are normalized (linearly re-scaled) to a scale from 0 to 1 such that 0 corresponds to the minimum response option and 1 corresponds to the maximum response option. As seen in this table, treatment group participants were more likely than those in the control group to stop watching their assigned video, yet most participants in both groups opted to watch the last five minutes. Meanwhile, treatment group participants reported a higher level of learning, a higher level of attention, and a greater degree of worldview change (both immediately after watching and at the one-week follow-up). Treatment group participants were also more accurate in their responses to the comprehension questions; however, the comprehension questions for the treatment video may have been easier (or harder) than the comprehension questions for the control video. Table 5Effects on other measures (besides consumption).Normalized mean valuep-value of differenceTreatmentControlStop watching0.400.15Comprehension0.840.70Sad0.770.06Angry0.590.04Bored0.110.57Disgusted0.810.13Surprised0.430.15Ashamed0.420.01Amount learned0.770.50Glad watched0.510.56Attention to video0.730.39Worldview change—part 10.460.16Worldview change—part 20.410.18To facilitate comparisons, all measures reported in the ‘treatment’ and ‘control’ columns of this table have been normalized (linearly re-scaled) to a scale from 0 to 1 such that 0 corresponds to the minimum response option and 1 corresponds to the maximum response option. The last column contains p-values associated with testing differences between the treatment and control normalized mean values. Effects on other measures (besides consumption). To facilitate comparisons, all measures reported in the ‘treatment’ and ‘control’ columns of this table have been normalized (linearly re-scaled) to a scale from 0 to 1 such that 0 corresponds to the minimum response option and 1 corresponds to the maximum response option. The last column contains p-values associated with testing differences between the treatment and control normalized mean values. Table 5 also reveals pronounced differences in participants’ emotional reactions to each video. Namely, treatment group participants expressed stronger emotional reactions for all of the emotions measured except boredom. Indeed, the control video elicited fairly high levels of boredom, but was otherwise quite limited in its emotional impacts. The treatment video elicited particularly high levels of sadness and disgust. In addition to our primary incentivized measure of consumption, we elicited a secondary self-reported measure of consumption. In particular, we asked participants, both before watching their assigned video and again in the part 2 follow-up survey, how many times in the last week they ate food in each of four food categories: mushrooms, chocolate, pork, and other meat besides pork (see Section 4 for additional details). In turn, Table 6 shows the effects of the videos on this secondary, self-reported consumption measure. These results should be interpreted with caution, in light of the many known limitations of such self-reported consumption measures. That said, it is arguably still interesting (if only as a curiosity) to measure these effects and compare them to the effects on our primary, incentivized consumption measure, especially considering the high reliance on self-reported consumption measures in the experimental literature. Table 6Self-reported consumption results. FoodCategoryGroupAvg. consump.Change fromPart 1 To 2p-valuePart 1Part 2PorkTreatmentControlT−COther meatTreatmentControlT−CTotal meat (pork + other)TreatmentControlT−CMushroomsTreatmentControlT−CChocolateTreatmentControlT−CThis table reports average consumption counts for each of various food categories in parts 1 and 2 of the study, and the change from part 1 to part 2, and p-values associated with the change values. For each food category, we report values for the treatment group, the control group, and the difference between the treatment and control groups ‘(T-C)’. The consumption measure is the (self-reported) number of times in the last week that the participant ate food in the indicated category. This measure was elicited using seven response options ranging from “0” to “14 or more.” In our analysis, responses specifying a range of values were converted to a single value using the midpoint, with the exception of “14 or more,” which was coded as 14. Self-reported consumption results. This table reports average consumption counts for each of various food categories in parts 1 and 2 of the study, and the change from part 1 to part 2, and p-values associated with the change values. For each food category, we report values for the treatment group, the control group, and the difference between the treatment and control groups ‘(T-C)’. The consumption measure is the (self-reported) number of times in the last week that the participant ate food in the indicated category. This measure was elicited using seven response options ranging from “0” to “14 or more.” In our analysis, responses specifying a range of values were converted to a single value using the midpoint, with the exception of “14 or more,” which was coded as 14. With these caveats in mind, Table 6 reveals statistically significant reductions in self-reported consumption of pork, of other meat besides pork, and of all meat among treatment group participants at the one-week follow-up. With that said, we see statistically significant reductions in self-reported consumption of pork, of other meat besides pork, and of all meat in the control group as well. While these effects are all larger in the treatment group than in the control group, only the difference in total self-reported meat consumption—amounting to a percent reduction in the treatment group and a percent reduction in the control group—is statistically significant. As for the other food categories, we observe significant increases in self-reported mushroom consumption after one week in both the treatment and control groups, with no changes in self-reported chocolate consumption. Although it is unclear why self-reported mushroom consumption would increase in both groups, it is unlikely that these increases were due to statistical noise as both effects are highly significant with . In any event, participants’ incentivized consumption choices do not reveal any changes in demand for mushrooms, as demand for the lone base meal that contains mushrooms (arugula salad) was unchanged after one week in both the treatment and control groups; that said, we did see higher demand for arugula salad immediately after watching the video. See Appendix Tables 9 and 10. Discussion In this study, we found that watching a roughly 15-minute clip from the Dominion documentary led to significant increases in demand for (more sustainable) meatless meals. Furthermore, this effect persisted at a one-week follow-up, which is noteworthy considering the lack of prior experimental evidence showing a persistent effect of watching a video on subsequent meat consumption. It is also noteworthy that we observe significant reductions in demand for chicken and beef—and not just for pork—even though our treatment video focuses on industrial-scale pork production. With that said, our treatment video still had a proportionally larger effect on demand for pork compared to other types of meat. This discrepancy, along with the relatively low baseline popularity of pork in our sample, especially compared to chicken, suggests that the impact of our treatment video may have been limited by its focus on pork. Our treatment video appears to have been quite effective in promoting plant-based eating compared to other videos tested in the literature. This could, in large part, reflect differences in video quality. For organizations that may consider the use of videos in promoting sustainable consumption, it may thus be crucial to be highly selective when choosing a video. As one specific possibility, using a video clip from the first 16 minutes of Dominion could be a sound strategy in light of its many awards and high viewer ratings, not to mention its effects on consumption as observed in our study. The effects of our treatment video may also be compared to the effects of many non-video interventions that have been tested in the literature, with mixed results37,40–48. (There are also studies that test multi-faceted interventions that include a video as one part49–51, also see related reviews and analysis52–56.) To give a sense of how the magnitude of our main effect compares to the measured effects of other interventions, we consult Green et al.’s56 meta-analysis of prior randomized controlled trials that measure the effects of various interventions on subsequent meat consumption (measured at least one day later). To facilitate comparisons, Green et al. express all effect sizes using the standardized mean difference (SMD). Across all studies included in the meta-analysis, the average SMD was .07, with a 95% prediction interval of −.11 to .24. In our study, the SMD for the effect of our treatment video on meat consumption after one week was 0.40, indicating a relatively large effect compared to the effects observed in other studies. Our study also demonstrates how emotional content may be effective in promoting sustainable consumption behavior. In this way, our study relates to a growing literature exploring how emotions may be leveraged to encourage environmentally-friendly behaviors57–60, an experimental economics literature testing the effects of induced or incidental emotions on pro-environmental behavior61–63, and a literature examining the relationship between emotions and dietary decisions64,65 including decisions to eat meat66. When considering the implications of our study, however, it is worth emphasizing that only one emotional response elicited by our treatment video—namely, shame—reliably predicted changes in associated consumption behavior. This suggests that other emotional responses, such as sadness, anger, and disgust, may impose unnecessary costs on the viewer. With that said, it is unclear whether an informative video like ours can induce a single emotion such as shame separately from other emotions. A more nuanced understanding of the relationship between negative emotional experiences and consumer well-being may also be useful in light of our finding that many participants who reported these emotions also indicated that they were glad to have watched the treatment video. Going forward, a key remaining challenge is to determine effective ways to encourage individuals to voluntarily watch such videos. After all, there is a well-documented tendency for consumers to avoid information about industrial-scale meat production24,31,32,67–70. With that said, it is noteworthy that most treatment group participants chose to continue watching our treatment video when given the option to skip the last 5 min. In future work, it could be useful to test whether our observed effects on consumption could be largely attained with a significantly shorter version of our treatment video, which may be more inviting to potential viewers and also improve compatibility with certain social media platforms such as TikTok and Instagram. It could also be useful to assess whether these effects can persist over longer time horizons and whether these effects would extend to participants’ real-world consumption choices in the field. As another direction, it could also be interesting to test the effects of repeated exposure, which could plausibly sensitize or de-sensitize viewers to the processes described in the video. Methods We performed a randomized controlled trial, in which participants were randomly assigned to a treatment group or a control group, which determined whether the participant was shown the treatment video segment from Dominion or an unrelated control video. Our study was pre-registered with the American Economic Association’s registry for randomized controlled trials (RCT ID AEARCTR-0014637, registered on 15/11/2024; see https://www.socialscienceregistry.org/trials/14637). This project received research ethics approval from the University of Toronto (protocol #46512). This study was conducted in accordance with relevant guidelines and regulations. Informed consent was obtained from all participants. Our study consisted of two parts. The first part was an in-person session held at the Rotman School of Management’s Behavioural Research Lab. During this session, participants completed a Qualtrics survey with their assigned video embedded in the survey. The second part was an online survey that was emailed to each participant exactly seven days after the participant completed the in-person session. This email included instructions to complete the part 2 survey within four days. An additional email reminding participants to complete the part 2 survey was sent three days later. Participants were recruited from the Rotman School of Management’s credit pool which is comprised of undergraduate students in introductory commerce courses who participate in research for course credit. To minimize attrition, participants were only awarded course credit after they completed both parts. The number of participants we were allotted in the semester in which we began our study was roughly half of our intended (registered) sample size. Consequently we ran our study in two consecutive semesters (Fall 2024 and Winter 2025). In total, 1,149 participants completed part 1 of our study. Of these, 1,111 participants also completed part 2. For our ex-ante power calculations, we assumed that 70 percent of participants would choose a meat-containing meal at baseline; we also assumed that 10 percent would choose a meatless meal at baseline then switch to a meat-containing meal after watching the video. To measure a net 5 percentage point effect size (i.e., a net reduction with 15 percent of participants choosing a meat-containing meal at baseline then switching to a meatless meal after watching the video) using a paired t-test to test with 95 percent confidence and 90 percent precision, a total sample size of 906 would be needed (assuming a 50-50 split between control and treatment groups). From this, we calculate a total target sample size of 1,208 under the assumption of a 25 percent attrition rate between parts 1 and 2 of our study. Naturally, our undergraduate student sample is not representative of the general population. We are therefore unable to say much about any potential impacts of our treatment video on other, non-student populations. That said, advocacy groups using similar interventions may seek to target young women (and may therefore be most interested in learning the effects of our treatment video on similar populations) in light of evidence of greater openness to plant-based eating among younger individuals71 and women37–39. Our undergraduate sample, in which female students are slightly over-represented (as seen in Table 1), may thus be somewhat aligned with the intended target audiences of our intervention if used in practice. Treatment and control videos For the treatment video, we used a 16-minute clip from the beginning of the Dominion documentary. While the full-length documentary addresses the use of various species of animals in a range of agricultural and other commercial applications, the segment in our treatment video focuses predominantly on industrial-scale pork production. We decided to use this particular clip based on anecdotal accounts suggesting the effects of watching Dominion are often experienced within the first 10 to 20 minutes, as well as our intention to use footage that focuses on a single species. This segment does not include any discussion of dietary choices, nor any appeals to adopt a plant-based diet. In this way, the task of connecting the experiences of pigs on industrial-scale farms (as depicted in the video) to one’s own consumption choices is left entirely to the viewer. In accordance with standard ethical guidelines, participants were informed at the start of the study that they could withdraw at any time without penalty. Additionally, after watching the first 11 minutes of the treatment video, participants were given an explicit choice of whether or not to watch the remaining five minutes. Participants who opted to stop watching still had to wait five minutes before proceeding to the post-video survey questions (with a blank screen displayed during this time and no permitted access to personal devices). Although the full documentary includes a detailed overview of methods used to slaughter pigs, only a portion of this content was included in our treatment video; furthermore, this portion was entirely contained in the last five minutes. Thus, participants who opted to stop watching after 11 minutes were not exposed to this content. Instead, these participants learned only about pigs’ experiences before they are sent to slaughter. For the control video, we used a 16-minute video clip from the documentary Life of the forest. Fungi, which addresses the role of mushrooms in forest ecosystems. While the full documentary includes some discussion of mushrooms as a food for human consumption, we excluded this content from our control video to avoid any potential influence on participants’ food choices. As with the treatment video, the control video featured a pause after the first 11 minutes, at which point participants were given a choice of whether to watch the remaining five minutes (while those who opted to stop watching were likewise shown a blank screen for the next 5 min). Primary (incentivized) consumption measure As our primary measure of consumption, we used an incentivized, lottery-based measure in which participants chose a menu item from a set of options available at a café located in the Rotman School of Management (the same building where the in-person sessions were held). This choice involved selecting a base meal and then selecting a protein to accompany the base. Of particular importance for our study, participants faced the same five protein options regardless of their chosen base: bacon, chicken, steak, tofu, and none. Although less important for our analysis, participants had four options for the base: bagel sandwich, asado bowl, arugula salad, and curry couscous. Participants chose a menu item at three points during the study: before watching the video, immediately after watching the video, and in the part 2 follow-up survey sent one week later. In our analysis, we examined the effects of the treatment video and the control video on participants’ protein choices, focusing mainly on the distinction between choosing a meat option versus a meatless option. Within each group, paired t-tests (two-sided) were used to test for a significant effect of watching the assigned video on subsequent protein choice compared to baseline choices. We also tested for a significant difference between treatment group effects and corresponding control group effects using independent two-sample t-tests (also two-sided). Participants had a roughly 50 percent chance of winning a voucher for one of the menu items that they selected in one of these three choices (lab rules prevented us from giving vouchers to all participants). At the end of the part 2 survey (and after making all three of these choices), participants were informed of whether or not they won a voucher. Those who won were then invited to pick up their voucher at the Rotman Behavioural Lab, with the designated meal printed on the voucher. Once picked up, the vouchers could then be redeemed at a café located one floor above the Rotman Behavioural Lab. It is important to acknowledge that our incentivized consumption measure was elicited in the context of a research study. It is therefore debatable whether or to what extent the measure would reflect participants’ private, real world consumption choices. Of note, however, Epperson and Gerster24 used a similar incentivized consumption measure, in which participants had a 50 percent chance of winning a voucher that could be redeemed for either a vegetarian or non-vegetarian meal at a local canteen, while also tracking participants’ actual food purchases at the canteen. The intervention used in the study had very similar effects on participants’ incentivized voucher choices in the lab and on actual choices in the field. While our particular measure is structured differently (in that we separately elicit choices for a base meal and an accompanying protein), Epperson and Gerster’s findings still lend credence to the external validity of an incentivized consumption measure in which participants have a roughly 50 percent chance of winning a voucher for a designated menu item. Our use of an incentivized measure as our primary measure of consumption (as opposed to a self-reported measure in which participants may lack the incentive to respond truthfully) was part of a broader effort to mitigate a potential social desirability bias—a major concern with self-reported consumption data72—whereby participants could be inclined to respond to survey questions in a manner that conforms with their perceptions of how the researchers want them to respond. Another feature of our effort to minimize social desirability bias was to include the following language in the first paragraph of our consent form: “A main purpose of the study is to assess whether watching a segment from a documentary video can impact beliefs and behavior. Prior research in this area has been inconclusive, with some studies suggesting ‘yes’ and other studies suggesting ‘no.’ In this study, we do not have a favored hypothesis. We are simply interested in learning the answer to the research question.” Here, our intent was to convey that we, the researchers, want only truthful answers. When interpreting a causal effect of the treatment video on our primary, incentivized consumption measure, however, it is important to keep in mind that some participants may still have felt as though their choices were “observed” by the researchers and that this feeling could plausibly sway choices in some cases. Other measures In addition to our primary, incentivized measure of consumption, we elicited a secondary, self-reported measure of consumption. In particular, we asked participants how many times in the last week they ate food in each of four food categories: mushrooms, chocolate, pork, and other meat besides pork. Participants were asked these questions before watching their assigned video and again in the part 2 follow-up survey. In each of these questions, participants could choose from seven options, ranging from “0” to “14 or more.” In our analysis, we converted responses specifying a range of values to a single value using the midpoint (e.g., coding “3 to 4” as 3.5), with the exception of the “14 or more” response option, which we coded as 14. Here, we included questions about mushroom and chocolate consumption to balance out our questions about meat consumption. In this way, it was arguably less clear to participants which consumption categories we were interested in studying, thus further mitigating a potential social desirability bias. In responses to an (optional) open-ended question asking participants what they believe our study is about, a number of participants speculated that we were interested in studying mushrooms or fungi more broadly. During the in-person session, we collected demographic information, asking participants to indicate their age, their sex, their racial or ethnic background, and their religion or creed. We also elicited a measure of participants’ “mindset flexibility” before they watched their assigned video. To do this, we included two questions, which were adapted from Minson et al.’s73 scale for measuring receptiveness to opposing views, while allowing participants to respond using a five-point Likert scale. To form our composite measure of mindset flexibility, we translated responses to numerical values from 1 to 5 and took the average of the responses to these two questions. After watching their assigned video, participants were asked four questions testing their comprehension of the video’s content (these questions covered content from only the first 11 minutes, prior to the pause where participants were given a choice to stop watching); we then used the percent of correct responses to these questions as a measure of a participant’s overall comprehension. Next, participants were asked to indicate, on a four-point scale, the extent to which they felt sad, angry, bored, disgusted, surprised, and ashamed when watching their assigned video. For example, the first of these six emotional response questions asked “Did you feel sad watching the video?” with “No,” “Yes, a little sad,” “Yes, moderately sad,” and “Yes, very sad” as the four response options (and similar for the other five emotions). Lastly, participants were asked how much they learned from the video (“a lot,” “a little,” or “nothing”), the extent to which they are glad or regretful that they watched the video (five-point scale, ranging from “very glad” to “very regretful”), their level of attention to the video (five-point scale, ranging from “very low” to “very high”), and how much their “worldview” changed from watching the video (six-point scale, ranging from “no change” to “very large change”). A version of this worldview question was also included in the part 2 follow-up survey. Acknowledgements We thank Seth Green and Mark Kamstra for providing invaluable feedback at multiple points during this study. We appreciate comments from participants who attended seminars hosted by the Stanford University Humane and Sustainable Food Lab and by the Animal Welfare Economics Working Group. We are grateful to Chris Delforce for allowing us to use a segment from the film Dominion and to Forest Film Studio and The State Forests of Poland for permission to use an excerpt of The life of the forest. Fungi in this study. We thank the Exchange Café at the University of Toronto for partnering with us on this study. All errors are our own. This project received research ethics approval from the University of Toronto (protocol #46512). This work was supported by an Insight Development Grant from the Social Sciences and Humanities Research Council of Canada. Appendix See Tables 7, 8, 9, 10, 11 and 12. Author contributions Both authors contributed equally to all aspects of this study. Data availability The (de-identified) data used in this study will be made available to any researcher who submits a methodologically sound proposal to peter.landry@utoronto.ca and signs a data access agreement. Additional information and materials related to this study are provided on our pre-registration page, https://www.socialscienceregistry.org/trials/14637 Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The (de-identified) data used in this study will be made available to any researcher who submits a methodologically sound proposal to peter.landry@utoronto.ca and signs a data access agreement. Additional information and materials related to this study are provided on our pre-registration page, https://www.socialscienceregistry.org/trials/14637