Logistic Regression Models in Sociological Research

Key takeaways

(file:///C:\Users\scott\Zotero\storage\S3L8QG52\Gayle_Lambert_2009_Logistic%20Regression%20Models%20in%20Sociological%20Research.pdf)

Bibliography: Gayle, V., Lambert, P.S., 2009. Logistic Regression Models in Sociological Research. DAMES Node, Technical Paper.

Authors:: Vernon Gayle, Paul S Lambert

Collections:: To Read

First-page: 18

Abstract

Many readers will be familiar with statistical modelling approaches to analysing social survey data. Statistical models offer sociologists an attractive method to summarize patterns in survey datasets (Dale and Davies, 1994; Goldthorpe, 2007). Sociologists have tended to employ regression models in order to explore the effects of multiple explanatory variables on an outcome of interest. Standard statistical modelling approaches are becoming increasingly widely known and in the UK sociology postgraduate students are routinely trained in these techniques.i Advances in software packages for statistical analysis (e.g. SPSS, 2008; Stata, 2007) and the rapid increases in the power of desktop computers have greatly expanded the possibilities for developing statistical models to analyse survey datasets. These advances have been coupled with increased accessibility to survey datasets, particularly through internet based links to national data archives.ii

Citations

content: "@gayleLogisticRegressionModels2009" -file:@gayleLogisticRegressionModels2009

Reading notes

Imported on 2024-05-07 21:16

⭐ Important

& In our experience the effects of the individual explanatory variables in logistic regression models are rarely well described in sociological studies (and habitually mystify sociology students). (p. 18)
& The result is that the β1 is the effect that a change in x1 has on the log odds of the outcome variable y taking the value 1. (p. 18)
& Moreover, for categorical explanatory variables the β associated with category effects should be thought of as the effect on the log odds of moving from the reference (or base) category to the particular category (or level) of the X variable (p. 18)
& The use of odds ratios are frequently advocated in methodological text books. Our overall position is that odds ratios should be avoided when interpreting the effect of explanatory variables and when communicating results from logistic regression models. (p. 18)
& If odds ratios must be used then terms like ‘higher’ and ‘lower’, and ‘increased’ and ‘decreased’, should be used to describe comparative effects (p. 21)