The evolving categories multinomial distribution: introduction with applications to movement ecology and vote transfer

Ricardo Carrizo Vergara, Marc Kéry, Trevor Hefley · 2025

Read The evolving categories multinomial distribution: introduction with applications to movement ecology and vote transfer by Ricardo Carrizo Vergara, Marc Kéry, Trevor Hefley (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-20151.txt.

Introduction

Ricardo Carrizo Vergara, Marc Kéry, and Trevor Hefley introduced a new probability distribution in twenty twenty-five for count data whose categories change over time. The evolving categories multinomial models individuals moving among groups. Groups can also appear, vanish, or change identity. They derive its one- and two-time marginals and its first and second moments.

With a Poisson prior on the unknown total population, a companion distribution follows. Since likelihoods are impractical, they propose two estimating functions for fitting. Applications show the range: inferring animal movement from surveys with shifting regions, tested on simulated wanderers and real lesser prairie-chickens. A second case estimates vote transfers between rounds of the twenty twenty-one Chilean presidential election.

Catalog record

CREATORRicardo Carrizo Vergara, Marc Kéry, Trevor Hefley
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE97 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-20151

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The evolving categories multinomial distribution: introduction with applications to movement ecology and vote transfer Ricardo Carrizo Vergara, Marc Kéry, Trevor Hefley Abstract Abstract We introduce the evolving categories multinomial (ECM) distribution for multivariate count data taken over time. This distribution models the counts of individuals following iid stochastic dynamics among categories, with the number and identity of the categories also evolving over time. We specify the one-time and two-times marginal distributions of the counts and the first and second order moments. When the total number of individuals is unknown, placing a Poisson prior on it yields a new distribution (ECM-Poisson), whose main properties we also describe. Since likelihoods are intractable or impractical, we propose two estimating functions for parameter estimation: a Gaussian pseudo-likelihood and a pairwise composite likelihood. We show two application scenarios: the inference of movement parameters of animals moving continuously in space-time with irregular survey regions, and the inference of vote transfer in two-rounds elections. We give three illustrations: a simulation study with Ornstein-Uhlenbeck moving individuals, paying special attention to the autocorrelation parameter; the inference of movement and behavior parameters of lesser prairie-chickens; and the estimation of vote transfer in the 2021 Chilean presidential election. Keywords Evolving categories multinomial (ECM) distribution, ECM-Poisson, composite likelihood, abundance modeling, movement ecology, ecological inference. 1 Introduction Count data are ubiquitous in statistics. Probability distributions for modeling integer-valued data have been developed since the very beginning of probability theory and are used in virtually every discipline that applies probability and statistics. One particular scenario often present in applications is when count data are taken over time. In addition, often counts refer to entities (e.g., particles, animals, people, or individuals for using a general term), whose attributes experience a dynamics over time. In this context, a certain multivariate distribution arises as a conceptually simple model for such counts: we call it here the evolving categories multinomial (ECM) distribution. It considers individuals moving from category to category over time, with the additional feature that the categories may also evolve over time . The categories are abstract and they may mean whatever is required in an application context where individuals have to be categorized and counted dynamically: e.g., spatial volumes for gas particles; survey landscapes for animal abundance; political preference, age group or professional activity sector of humans; seasonal consumption goods preferred by costumers; health state of agents during pandemics, to mention but a few examples. The ECM distribution is the joint distribution followed by the counts per category and time of individuals following independently the same dynamics across the evolving categories. It can be used to infer movement properties using the counts as sole information. While the construction of this multivariate discrete distribution is simple and general, we are unaware of any work focusing on its general abstract form or mathematical properties. Technically, any application with counts of independently moving individuals has used this probabilistic structure, so in that sense the distribution is not new. Our contribution is to pay special attention to this mathematical object, giving it a name, studying its basic properties, developing methods for statistical inference and demonstrating its utility by addressing important questions in scientific fields such as ecology and sociology. One important aspect of this abstract construction is that the exact individual “position” at each time is lost, and an aggregate becomes the sole available information. This key issue is present in every application where the individual information is unavailable because it is too expensive, sensitive or even physically impossible to collect. The classical example of statistical mechanics ( Gallavotti, (2013) ; Pathria, (2017) ) for studying broad scale properties of many-particles systems is perhaps the most common example. In the social sciences, some aggregate information about people’s behavior can be available for some institutions (e.g., market and educational institutions, social media, governments) or even publicly available (e.g., vote counts in transparent democracies), but sensitive individual information is often protected. In ecology, both kinds of information are often available: animal count data from surveys ( Seber, (1986) ) and telemetry data of animal’s positions from tracking devices ( Hooten et al., (2017) ) . But tracking animals has inconveniences such as biased sampling or expensive tracking technology, while count data are very widely available, usually cheaper to collect, and can be obtained from citizen-science data collection initiatives ( Sullivan et al., (2009) ) . In this work we study the minimal mathematical properties of this distribution so we can apply it to model important phenomena and provide fitting techniques. We aim for two application scenarios from ecology and sociology. In the ecological scenario, we infer animal movement parameters from survey data (spatio-temporal counts) taken at irregular times in continuous space-time. There is a substantial literature on the modeling of animal movement ( Turchin, (1998) ; Grimm & Railsback, (2013) ) and of abundance ( Andrewartha & Birch, (1954) ; Thorson & Kristensen, (2024) ) separately, but the two concepts have hardly ever been united in a single framework, either in practice with integrated data scenarios or in theory by studying the statistical properties issued from the conceptual connection between movement and count. We refer to ( Chandler et al., (2022) ; Roques et al., (2022) ; Potts & Börger, (2023) ; Buderman et al., (2025) ) for some recent works using, either explicitly or implicitly, the connection between movement and broad-scale count. An important and hitherto neglected aspect is the space-time autocorrelation induced by movement. Here we exploit it to infer movement parameters which do not occur on the average (intensity) field. We illustrate with a simulation study with steady-state Ornstein-Uhlenbeck moving individuals. Also, we show how the ECM distribution is useful in classification problems, where the proportions of animals following different movement behaviors must be estimated. We illustrate with an analysis of translocated lesser prairie-chickens ( Tympanuchus pallidicinctus ) in the U.S. ( Berigan et al., (2024) ) . In the sociological scenario, we study vote transfer between election rounds, which is an ecological inference problem ( Schuessler, (1999) ; King et al., (2004) ) . Here, the ECM distribution offers a different approach to existing methodology commonly found in the literature. We illustrate by inferring vote transfer during the 2021 Chilean presidential election. Our work is organized as follows. In Section 2 we introduce the ECM distribution with its main properties: characteristic function, one-time and two-times marginal distributions, and first and second order moments. In Section 3 we Poissonize the ECM distribution in order to consider cases with unknown total number of individuals, describing thus a new distribution here called ECM-Poisson. Since likelihoods are intractable, in Section 4 we propose two ad-hoc estimating functions for inference which allow the estimation of parameters occurring in the mean and the autocorrelation of the counts. The first is a Gaussian pseudo-likelihood respecting mean and covariance, justified by the central limit theorem and conceived for cases with huge number of individuals. The second is a pairwise composite likelihood, conceived for cases where the number of individuals is too low for the Gaussian pseudo-likelihood to be reliable. The distributions of the pairs are related to the bivariate binomial and Poisson distributions ( Kawamura, (1973) ) . Applications are shown in Section 5 .

Read the complete text