|LDA||QV - Online Calculator||RDI|
This page features example SPSS and STATA command files illustrating implementations of Quasi-Variance calculations for social science example applications. It is maintained by Vernon Gayle and Paul Lambert.
Background / Quasi-Variance tools / Further links
These pages refer to a draft paper by Vernon Gayle and Paul Lambert which introduces Quasi-Variance methods to a sociological audience
Gayle, V and Lambert, PS (2006) Using Quasi-Variance to communicate sociological results from statistical models(long version). Stirling: University of Stirling, Working paper 2006-3 of the Researcher Development Initiative project 'Longitudinal Data Analysis for Social Science Researchers'.
Gayle, V. and Lambert, PS (2007) Using Quasi-Variance to communicate sociological results from statistical models, Sociology, 41(6), 1191-1206.
'Quasi-Variances' are statistics associated with the parameter estimates of the different levels of categorical explanatory variables within regression format statistical models. The paper recommends that researchers publishing regression model results should routinely seek to present appropriate Quasi-Variance statistics, since these enable readers of these outputs to readily perform tests of difference between any combination of categorical factor parameter estimates (not otherwise possible without access to the full variance-covariance matrix for the estimates).
Quasi-Variance methods were introduced by David Firth. He has produced an online web-calculator which can be used by reseachers who produce models, in order to generate appropriate Quasi-Variance statistics.
When Quasi-Variance statistics have been generated, they would ordinarily be used in three ways:
- They may be reported next to the appropriate parameter estimates in a regression model output
- They may be used to undertake Wald-tests for the significance of the difference between any two categories of the categorical variable : this may be achieved by entering the unstandardised parameter coefficient and quasi-variance into this MS-Excel calculator
- They may be used to construct approximate confidence intervals for the differences between all categories of the categorical variable : these may also be generated by entering the unstandardised parameter coefficient and quasi-variance into this MS-Excel calculator
AN EXAMPLE (Figure 1 in the paper):
The links below lead to access to example command files in SPSS and STATA which generate regression model results and corresponding data for the Quasi-Variance calculation following the examples from the associated paper.
STATA: examples and data SPSS: examples [to follow] and data Example 1: Logistic regression using 2001 UK Census SARs qv_example_1.do [link to data] qv_example_1.sps [link to data] Example 2: Linear regression using 2002 UK General Household Survey extract qv_example_2.do ghs2002.dta qv_example_2.sps ghs2002.sav Example 3: Logistic regression with interactions, 2002 UK General Household Survey extract qv_example_3.do ghs2002.dta qv_example_3.sps ghs2002.sav Example 4: Logistic regression panel model using 1986 SCELI employment data extract qv_example_4.do wemp.dta n/a wemp.dat Macros: (these STATA / SPSS programmes are used within the example files above) contrasts_tests.do (runs a sequence of Wald tests - compares categories on the original data) makedummyvars.sps (SPSS utility for creating dummy / indicator variables) covmat.sps (SPSS ultility for generating variance-covariance matrix of variables)
(referred to within command files)
(Note: the *.do and *.sps command files are released here as plain text files for ease of browsing. To download them and then run, you will need to change the name of the file, for instance file1_do.txt should be renamed as file1.do).
This site is connected with the ESRC funded training project 'Longitudinal Data Analysis for Social Science Reseachers', which is part of the ESRC's Researcher Development Initiative.
Last modified 3 July 2006
This document is maintained by Paul Lambert (firstname.lastname@example.org) and Vernon Gayle (email@example.com)