A command to estimate and interpret models of dynamic compositional dependent variables: New features for dynsimpie

Abstract

Philips, Rutherford, and Whitten (2016, Stata Journal 16: 662–677) introduced dynsimpie, a command to examine dynamic compositional dependent variables. In this article, we present an update to dynsimpie and three new adofiles: cfbplot, effectsplot, and dynsimpiecoef. These updates greatly enhance the range of models that can be estimated and the ways in which model results can now be presented. The command dynsimpie has been updated so that users can obtain both prediction plots and change-from-baseline plots using postestimation commands. With the new command dynsimpiecoef, various types of coefficient plots can also be obtained. We illustrate these improvements using monthly data on support for political parties in the United Kingdom.

Policy agenda trade‐offs for sustainability: The compositional change of attention about energy in legislative hearings

“…Dynamic compositional models as the main method of empirical investigation can be a useful way to analyze possible trade‐offs over time because trade‐offs between policy areas can be empirically examined better by understanding that there are limited numbers and degrees of political endeavors and policies that can be devoted at a given time (Jung et al., 2020; Philips et al., 2016). By considering the aggregate of all political agendas in a given time (such as a year) as compositional variables—variables that are made up of values that sum to 100% for each unit of analysis—the dynamic compositional model shows how relative compositions change over time.…”

“…In all figures, it is assumed that there is a shock, a 1‐standard‐deviation increase in unemployment (which is around 1.602). Moreover, the influence of this shock will be examined through the change‐from‐baseline plots of simulated output and the expected proportion change plots from estimated results, respectively (Jung et al., 2020, p. 589). Especially in Figures 2 and 4, another economic shock, proxied with a 1‐standard‐deviation decrease in percent change in GDP (which is around −2.133), will be added along with a 1‐standard‐deviation increase in unemployment.…”

“…Focusing on trade‐offs between policy domains can be empirically examined when we begin to consider the limited number of political endeavors and policies that can be devoted at a given time and how these change over time. In this regard, the dynamic compositional model (Jung et al., 2020; Philips et al., 2016) can be used as the main method of empirical investigation for this article and related projects by considering that the aggregate of all political agendas, budget allocations (Adolph et al., 2020; Philips et al., 2016), and policies and laws in a specific period of time (such as a year) can be expressed as compositional variables—variables that are made up of values that sum up to 100% for each unit of analysis.…”

Welfare state regimes and social policy resistance to fiscal consolidations

“…We use the dynsimpie package in Stata developed by Jung et al (2020) for compositional dependent variable analysis. Dynsimpie does a log transformation of the data to free them of the constraint of summing to zero, making them unbounded and independent, so that conventional linear techniques can be used.…”

A careful consideration of CLARIFY: simulation-induced bias in point estimates of quantities of interest

“… 1 CLARIFY for Stata (Tomz et al , 2003), Zelig for R (Imai et al , 2008; Choirat et al , 2018), and clarify for R (Greifer et al 2023) simulate quantities of interest and find the average. The package dynsimpie for Stata also reports the average of simulations (Philips et al , 2016; Jung et al , 2020). The margins command in Stata (StataCorp, 2017), the margins package in R (Leeper, 2021), and the predict() function in R for the glm (R Core Team, 2018) and polr (Venables and Ripley, 2002) classes directly transform coefficients into the quantities of interest.…”