SOCI209 - DESCRIPTION OF FINAL

The final for SOCI 209 will have between 42 and 50 multiple-choice questions.  The following list of topics covers at least 85 percent of the questions on the final.  A few questions will be added between now (25 Apr) and the day of the final (2 May).  Corresponding topics will be added to this list as soon as available.

• given a standard multiple regression printout with some entries blanked out, know how to reconstruct
• the multiple r
• the squared multiple r (R-square)
• the adjusted squared multiple r
• the df for the regression
• the df for the error (aka residual)
• the F-ratio for the regression
• the t-ratio for testing that an individual regression coefficient is different from zero
• the appropriate t-test for a regression coefficient
• VIF knowing TOL and vice-versa
• apply the usual rule of thumb for TOL or VIF
• given the necessary information (such as SSE(F) and SSE(R)) test whether several regression coefficients are simultaneously equal to zero using comparison of full versus reduced model and the appropriate F-test (2 questions)
• in which kind of research situation (e.g., controlled experiment, exploratory observational study, etc.) is model specification (i.e., choosing which independent variables to include in the model) likely to be most challenging?
• in the all-possible-regressions method of selecting the regression model, what are the different approaches to selecting a subset of independent variables to include? (2 questions)
• what is a major weakness of automatic search procedures for model selection, such as forward stepwise regression?
• how does one construct a partial regression plot (a.k.a added-variable plot or adjusted variable plot)
• given a regression printout and the diagnostics for outliers and influential cases, know
• how to interpret the meaning of TOL/VIF (requires understanding how TOL/VIF is calculated)
• what constitutes evidence of collinearity
• what constitutes evidence of heteroskedasticity
• the interpretation of leverage (requires knowing what the sum of h_ii is)
• how to flag a case that's outlying with respect to the Y-dimension
• how to flag a case that's outlying with respect to the X-dimensions
• how to flag a case that's influential (using the standard rule of thumb)
• in the context of robust regression what is the meaning of MAD
• what method(s) exist for robust outlier detection in the presence of several outliers
• the consequences of collinearity for model estimation
• properties of remedial measures for collinearity, and in particular what estimator alleviates collinearity by sacrificing unbiasedness in favor of reducing estimator variance
• how to detect heteroskedasticity in a plot of residuals against predictor
• the principle of the Breusch-Pagan test of heteroskedasticity
• properties of alternative tests for heteroskedasticity, in particular which test of heteroskedasticity is robust even in the presence of non-normality of the errors
• consequences of heteroskedasticity for model estimation
• which robust (i.e., HCCM) estimator of standard errors has the greatest affinity with the concept of the deleted (a.k.a external) residual?
• what is the general strategy recommended by Long and Ervin (2002) when heteroskedasticity is suspected?
• the reason(s) why it is generally better to use an HCCM estimator rather than WLS when heteroskedasticity is suspected
• in the context of the bootstrap what is the distinction between fixed-X and random-X sampling
• in the context of the bootstrap, in what circumstances does one use  fixed-X or random-X sampling
• how to detect serial correlation of errors in time series data by looking at a plot of residuals against time

• what substantive process(es) might produce serially correlated errors in time series data
• consequences of serial correlation of errors for model estimation
• interpretation of the first-order autoregressive model
• properties of remedial measures for serial correlation of the errors
• doing a Durbin-Watson test given the D-W statistic