- Practical skills in data handling, scale construction and experience with statistical analyses for social science research;
- Knowledge of these assumptions underlying the statistical models (generalized linear models),
- Skills in evaluating whether the assumptions are reasonably met in a research problem,
- Ability to translate between the research problem and statistical models,
- Ability to select appropriate models among the familiar models;
- Ability to report analysis, both in statistical and substantive terms.
The first part of the course involves data handling, scale construction (mainly classical test theory, factor analysis), and the disciplined cautious use of syntax in statistical software (MERM: SPSS; SaSR: Stata). The second part of the course addresses statistical models, mostly from the class of generalized linear models. Models to be discussed:
(1) Linear regression modeling for continuous dependent variables (such as income), focusing on modeling mediation, interactions and nonlinear relations;
(2) Regression models for binary dependent variables (such as whether or not people are employed): logistic and probit regression, emphasizing different types of interpretaions;
(3) Regression models for ordinal dependent variables (e.g., voting intentions measured by a Likert item): ordinal logit.
(4) Regression models for nominal dependent variables with subject-level and/or alternative-level predictors (e.g., the denomination of the school attended by children): multinomial logit and conditional logit models;
(5) Regression models for the time-until-the-occurrence-of-an-event (e.g., the birth of the first child, the divorce of a marriage, promotion in a career): discrete time survival analysis and Cox regression.
The discussion on these regression-type models focuses on the substantive interpretation of these non-lineair models. In addition, we discuss numerous general statistical issues such as statistical estimation methods; Wald testing versus likelihood ratio testing; Hausman-style tests for model specification; marginal effects, etc. Computer practical for MERM (using SPSS) and SaSR (using Stata) are separate.