Produces two-stage (or slopes as outcomes) results for linear and logistic regr > ession models (beta version)
mlt2stage yvar xvar(s) [if] [weight] , l2id(varname) [ drop ] [ logit ] [ taboff ]
mlt2stage is part of the mlt (multilevel tools) package.
mlt2stage is an easy way to produce two-stage results. It calculates separate linear or logit regression models on yvar for each level2-unit l2id(varname), displays the results as table and stores the coefficients of the independent variables (xvar(s)). The logit option allows calculating logistic regression models.
mlt2stage can be used together with mltl2scatter to produce two-stage plots of the estimated single country regression coefficients of a lower-level variable over a higher-level variable. See Mood (2010) for the comparison of logit models based on different samples.
mlt2stage allows to specify a weight for the units at the lower level. aweights, fweights and iweights are allowed. See the help for regress and logit to read how these weights are treated.
drop coefficients are not stored, mlt2stage only produces an output table.
logit calculates logistic instead of linear regression models.
taboff no output table is shown, mlt2stage only stores the coefficients. This option is recommended if a large number of xvars is specified.
A simple two-stage plot
Load data set (ISSP 2006) . webuse redistribution.dta
Regress "Support for income redistribution" on age and sex . mlt2stage gr_incdiff age sex, l2id(Country)
Scatter plot showing the association between the (age- and sex-adjusted) level of support for redistribution and economic inequality (gini) . mltl2scatter cons_gr_incdiff gini, l2id(Country) labels qfit
Using mlt2stage for a graphic inspection of a multilevel model with cross-level interactions
Load data set (ISSP 2006) . net get mlt . use redistribution.dta
A multilevel regression of "Support for income redistribution" on income, age, gender and economic inequality (gini) . xtmixed gr_incdiff incperc age sex gini ia_gini_incperc || Country: incperc, mle var cov(un)
The model gives a significant interaction effect between the country-level variable economic inequality (gini) and the individual-level variable income (cross-level interaction). The model suggests that the negative effect of income becomes weaker if inequality is higher. Is this a robust result? Let's use mlt2stage!
Regress "Support for income redistribution" on all individual-level variables (income, age and sex) . mlt2stage gr_incdiff incperc age sex, l2id(Country)
Scatter plot showing the association between the slope of income and the country-level variable economic inequality (gini) . mltl2scatter coef_gr_incdiff_incperc gini, l2id(Country) labels qfit
By plotting the estimated slopes of income against the country-level variable we can visualize the interaction effect. It seems that there is a particular country which is responsible for the positive interaction effect estimated in the multilevel model. This country is Chile (Country == 152).
Scatter plot showing the association between the slope of income and the country-level variable economic inequality, Chile excluded (gini) . mltl2scatter coef_gr_incdiff_incperc gini if Country != 152, l2id(Country) labels qfit
Re-estimate the multilevel regression without Chile . xtmixed gr_incdiff incperc age sex gini ia_gini_incperc || Country: incperc if Country != 152, mle var cov(un)
The interaction effect is no longer significant. It was actually due to one particular Country.
ISSP (2006): International Social Survey Programme - Role of Government IV, GESIS StudyNo: ZA4700, Edition 1.0, doi:10.4232/1.4700.
Carina Mood (2010): “Logistic Regression: Why We Cannot Do What We Think We Can Do, and What We Can Do About It.” European Sociological Review 26 (1): 67-82.
Katja Moehring, GK SOLCIFE, University of Cologne, email@example.com, www.katjamoehring.de.
Alexander Schmidt, GK SOCLIFE and Chair for Empirical Economic and Social Research, University of Cologne, firstname.lastname@example.org, www.alexanderwschmidt.de.