Determining relative importance in Stata using dominance analysis: domin and domme

2021 ◽  
Vol 21 (2) ◽  
pp. 510-538
Author(s):  
Joseph N. Luchman
Keyword(s):  
Statistical Models ◽  
Common Method ◽  
Parameter Estimates ◽  
Relative Importance ◽  
Dominance Analysis ◽  
Independent Variables ◽  
Data Analytic

Dominance analysis is a common method applied to statistical models to determine the importance of independent variables. In this article, I describe two community-contributed commands, domin and domme, that can be used to dominance-analyze both independent variables and parameter estimates in Stata estimation commands. I discuss how to compute dominance statistics, provide multiple examples of each command applied to data, and outline how to interpret the results from each data-analytic example. I conclude with computational considerations for users applying larger models.

scholarly journals Relative importance analysis with multivariate models: Shifting the focus from independent variables to parameter estimates

2020 ◽  
Vol 4 (2) ◽  
pp. 1-20
Author(s):  
Joseph N. Luchman ◽  
Xue Lei ◽  
Seth Kaplan
Keyword(s):  
Statistical Models ◽  
Likelihood Estimation ◽  
Parameter Estimate ◽  
Theory And Practice ◽  
Parameter Estimates ◽  
Relative Importance ◽  
Multivariate Models ◽  
Independent Variables ◽  
Dependent Variables ◽  
Independent Variable

Conclusions regarding the relative importance of different independent variables in a statistical model have meaningful implications for theory and practice. However, methods for determining relative importance have yet to extend beyond statistical models with a single dependent variable and a limited set of multivariate models. To accommodate multivariate models, the current work proposes shifting away from the concept of independent variable relative importance toward that of parameter estimate relative importance (PERI). This paper illustrates the PERI approach by comparing it to the evaluation of regression slopes and independent variable relative importance (IVRI) statistics to show the interpretive and methodological advantages of the new concept and associated methods. PERI’s advantages above standardized slopes stem from the same fit metric that is used to compute PERI statistics; this makes them more comparable to one another than standardized slopes. PERI’s advantages over IVRI stem from situations where independent variables do not predict all dependent variables; hence, PERI permits importance determination in situations where independent variables are nested in dependent variables they predict. We also provide recommendations for implementing PERI using dominance analysis with statistical models that can be estimated with maximum likelihood estimation combined with a series of model constraints using two examples.

The Relative Importance of Socioeconomic and Political Variables for Public Policy

1977 ◽  
Vol 71 (2) ◽  
pp. 559-566 ◽  
Author(s):  
Michael S. Lewis-Beck
Keyword(s):  
Public Policy ◽  
Path Analysis ◽  
Multiple Regression ◽  
Partial Correlation ◽  
Policy Outcomes ◽  
Statistical Techniques ◽  
Relative Importance ◽  
Simple Correlation ◽  
Independent Variables ◽  
Political Variables

Since Dawson and Robinson, a dominant issue in the quantitative study of public policy has been the relative importance of socioeconomic and political variables for determining policy outcomes. It is argued here that past efforts to resolve this issue have been unsatisfactory, largely because they relied on inadequate statistical techniques, i.e., simple correlation, partial correlation, or multiple regression. Coefficients from these techniques are irrelevant for all but the most peculiar models of public policy. In general, if the researcher wishes to assess the relative importance of independent variables, it will be necessary to resort to path analysis of a formally constructed causal model. The comparison of “effects coefficients,” derived from path analysis, is offered as the preferred means of evaluating independent variables, superior to comparisons of coefficients from simple correlation, partial correlation, or multiple regression. When the effects coefficients are actually calculated for a popular model of welfare policy, socioeconomic variables appear much more important than political variables, contrary to interpretations coming from the more traditional statistical techniques.

scholarly journals A QUASI-LIKELIHOOD APPROACH TO PARAMETER ESTIMATION FOR SIMULATABLE STATISTICAL MODELS

2014 ◽  
Vol 33 (2) ◽  
pp. 107 ◽  
Author(s):  
Markus Baaske ◽  
Felix Ballani ◽  
Karl Gerald Van den Boogaart
Keyword(s):  
Parameter Estimation ◽  
Statistical Models ◽  
Estimation Method ◽  
Score Function ◽  
Boolean Model ◽  
Parameter Estimates ◽  
Random Errors ◽  
Likelihood Functions ◽  
Simulation Based ◽  
Likelihood Approach

This paper introduces a parameter estimation method for a general class of statistical models. The method exclusively relies on the possibility to conduct simulations for the construction of interpolation-based metamodels of informative empirical characteristics and some subjectively chosen correlation structure of the underlying spatial random process. In the absence of likelihood functions for such statistical models, which is often the case in stochastic geometric modelling, the idea is to follow a quasi-likelihood (QL) approach to construct an optimal estimating function surrogate based on a set of interpolated summary statistics. Solving these estimating equations one can account for both the random errors due to simulations and the uncertainty about the meta-models. Thus, putting the QL approach to parameter estimation into a stochastic simulation setting the proposed method essentially consists of finding roots to a sequence of approximating quasiscore functions. As a simple demonstrating example, the proposed method is applied to a special parameter estimation problem of a planar Boolean model with discs. Here, the quasi-score function has a half-analytical, numerically tractable representation and allows for the comparison of the model parameter estimates found by the simulation-based method and obtained from solving the exact quasi-score equations.

scholarly journals International migration beyond gravity: A statistical model for use in population projections

2008 ◽  
Vol 105 (40) ◽  
pp. 15269-15274 ◽  
Author(s):  
Joel E. Cohen ◽  
Marta Roig ◽  
Daniel C. Reuman ◽  
Cai GoGwilt
Keyword(s):  
International Migration ◽  
Generalized Linear Model ◽  
Bayesian Information Criterion ◽  
Information Criterion ◽  
Population Projections ◽  
Parameter Estimates ◽  
Independent Variables ◽  
Indicator Variables ◽  
Starting Model ◽  
Country Specific

International migration will play an increasing role in the demographic future of most nations if fertility continues to decline globally. We developed an algorithm to project future numbers of international migrants from any country or region to any other. The proposed generalized linear model (GLM) used geographic and demographic independent variables only (the population and area of origins and destinations of migrants, the distance between origin and destination, the calendar year, and indicator variables to quantify nonrandom characteristics of individual countries). The dependent variable, yearly numbers of migrants, was quantified by 43653 reports from 11 countries of migration from 228 origins and to 195 destinations during 1960–2004. The final GLM based on all data was selected by the Bayesian information criterion. The number of migrants per year from origin to destination was proportional to (population of origin)0.86(area of origin)−0.21(population of destination)0.36(distance)−0.97, multiplied by functions of year and country-specific indicator variables. The number of emigrants from an origin depended on both its population and its population density. For a variable initial year and a fixed terminal year 2004, the parameter estimates appeared stable. Multiple R2, the fraction of variation in log numbers of migrants accounted for by the starting model, improved gradually with recentness of the data: R2 = 0.57 for data from 1960 to 2004, R2 = 0.59 for 1985–2004, R2 = 0.61 for 1995–2004, and R2 = 0.64 for 2000–2004. The migration estimates generated by the model may be embedded in deterministic or stochastic population projections.

scholarly journals Using Dominance Analysis to Decompose Narcissism and Its Relation to Aggression and Externalizing Outcomes

Assessment ◽  
2017 ◽  
Vol 26 (2) ◽  
pp. 260-270 ◽  
Author(s):  
Colin E. Vize ◽  
Katherine L. Collison ◽  
Michael L. Crowe ◽  
W. Keith Campbell ◽  
Joshua D. Miller ◽  
...  
Keyword(s):  
Externalizing Behavior ◽  
Factor Model ◽  
Five Factor Model ◽  
Self Esteem ◽  
Levels Of Analysis ◽  
Relative Importance ◽  
Dominance Analysis ◽  
Multidimensional Construct ◽  
Trait Level ◽  
Multiple Levels

Research on narcissism has shown it to be multidimensional construct. As such, the relations the larger construct bear with certain outcomes may mask heterogeneity apparent at the more basic trait level. This article used the Five Factor Narcissism Inventory, a Five-Factor Model–based measure of narcissism that allows for multiple levels of analysis, to examine the relative importance of narcissistic traits in relation to aggression, externalizing behavior, and self-esteem outcomes in two independent samples. The relative importance of the narcissism factors was determined through the use of dominance analysis—a relatively underused method for determining relative importance among a set of related predictors. The results showed that antagonism, compared with agentic extraversion and neuroticism, was the dominant predictor across all forms of aggressive behavior. Additional analyses showed that subscales within the broader factor of antagonism also showed differential importance relative to one another for certain aggression outcomes. The results are discussed in the context of the relation between narcissism and aggression and highlight the utility of using extensions of regression-based analyses to explore the heterogeneity within personality constructs.

Bias in the Estimation of Functional Relationships from Time Series Data

1985 ◽  
Vol 42 (1) ◽  
pp. 147-149 ◽  
Author(s):  
Carl J. Walters
Keyword(s):  
Time Series ◽  
Measurement Errors ◽  
Time Series Data ◽  
Series Data ◽  
Parameter Estimates ◽  
Natural Experiments ◽  
Independent Variables ◽  
Functional Relationships ◽  
Stock Recruitment ◽  
The Relationship

Functional relationships, such as stock–recruitment curves, are generally estimated from time series data where natural "random" factors have generated both deviations from the relationship and also informative variation in the independent variables. Even in the absence of measurement errors, such natural experiments can lead to severely biased parameter estimates. For stock–recruitment models, the bias is misleading for management: the stock will appear too productive when it is low, and too unproductive when it is large. The likely magnitude of such biases can and should be determined for any particular case by Monte Carlo simulations.

scholarly journals Modelling annual mass balances of eight Scandinavian glaciers using statistical models

2015 ◽  
Vol 9 (1) ◽  
pp. 383-415 ◽  
Author(s):  
M. Trachsel ◽  
A. Nesje
Keyword(s):  
North Atlantic ◽  
Statistical Models ◽  
Atlantic Multidecadal Oscillation ◽  
Relative Importance ◽  
Mass Balances ◽  
Average Mass ◽  
The North ◽  
Nao Index ◽  
Time Period ◽  
The North Atlantic Oscillation

Abstract. Glacier mass balances are mainly influenced by accumulation-season precipitation and ablation-season temperature. We use a suite of statistical models to determine the influence of accumulation-season precipitation and ablation-season temperature on annual mass balances of eight Scandinavian glaciers, ranging from near coastal, maritime glaciers to inland, continental glaciers. Accumulation-season precipitation is more important for maritime glaciers, whereas ablation-season temperature is more important for annual balances of continental glaciers. However, the importances are not stable in time. For instance, accumulation-season precipitation is more important than ablation-season temperature for all glaciers in the 30 year period 1968–1997. In this time period the Atlantic Multidecadal Oscillation (AMO) index was consistently negative and the North Atlantic Oscillation (NAO) Index was consistently positive between 1987 and 1995, both being favourable for glacier growth. Hence, the relative importance of precipitation and temperature for mass balances is possibly influenced by the AMO and the NAO. Climate sensitivities estimated by statistical models are similar to climate sensitivities based on degree-day models, but are lower than climate sensitivities of energy balance models. Hence, future projections of mass balances found with our models seem rather optimistic. Still, all average mass balances found for the years 2050 and 2100 are negative.

Sign in / Sign up

Export Citation Format

Share Document