Stepwise Regressions in Language Variation Suite – LVS

LVS provides three types of model comparison (LRT, AIC, and BIC) using the package MASS. The stepwise regression uses both directions (step up and step down) and selects the best model (best predictors).

All three criteria assess model fit.  LRT is based on log likelihood ratio (k = qchisq(1-p, df=1), where for p=0.05, k = 3.84). For more information on AIC ( Akaike Information Criterion ) and BIC (Bayesian information criterion) – see  http://www.jmp.com/support/help/Likelihood_AICc_and_BIC.shtml.

Steps to perform stepwise regression in LVS:

  1. Upload csv or excel file – Panel DATA Screen shot 2016-08-06 at 5.04.42 PM
  2. Go to Panel INFERENTIAL STATISTICS – tab MODELINGScreen shot 2016-08-06 at 5.06.33 PM
  3. Select your regression model (dependent and independent factors), type of regression (see tab REGRESSION) and click RUN regression.
  4. Go to STEPWISE REGRESSION tab and click RUN stepwise model.Screen shot 2016-08-06 at 5.11.55 PM
  5. Return to Modeling and Regression and update your selection with the best fitted model.

As always, your feedback and suggestions are greatly appreciated! (LVS Team)

Advertisements

Varbrul Weights in LVS

Language Variation Suite has added a Varbrul analysis. Varbrul is “an implementation of logistic regression that is used by many sociolinguists” (Keith Johnson, 2008, 174). At present LVS calculates Varbrul weights for a binary dependent variable and categorical independent variables. Varbrul output format is based on chapter 5.7 (K.Johnson,2008):  inverse logit (inv.logit) is used from the package gtools. Contrasts option in logistic regression is set to contr.sum. For a binary variable, the calculation is inv.logit(coeficient*1) and inv.logit(coeficient*-1), which outputs weights for two values (e.g. men and women).

Screen shot 2016-08-04 at 12.18.48 AM

Shiny application allows for an interactive quantitative analysis and it is based on R programming language. Since this toolkit is still under development, we will greatly appreciate its evaluation, comments, and feedback!