Replication data for

Electoral Institutions and the Politics of Coalitions:

Why Some Democracies Redistribute More Than Others

The analysis, and data, fall into two parts: One uses redistribution as a dependent variable, the other on government partisanship. If you use the data please use this citation:

Torben Iversen and David Soskice, “Electoral Institutions and the Politics of Coalitions: Why Some Democracies Redistribute More Than Others.” American Political Science Review 100 (2), 2006: 165-181.

I. Redistribution (get Stata dataset here):

The regression is based on an error correction model, adjusting for unequally spaced observations on the dependent variable. The exact procedure is described on p. 174 of the article.

For each variable in the data there are 10 different iterations, one for each value of ρ (or “Rho” as it is referred to in the data) between 0 and .9 (using .1 increments).

The dependent variable, R, is redistribution (measured as the percentage reduction in the gini from before to after taxes and transfers) minus Rho times the LDV (see formula on p. 174). R07, for example, is the dependent variable when Rho=7.

The independent variables are cumulative measures for different values of Rho and given the gap in years on the dependent variable. The cumulative variables have been computed in a separate spreadsheet (not included here).

The estimation of the model (reported in Table 5 on p. 175) was originally done in Stata6 using the following Stata gls procedures:

xtgls R07 p905007 stcogcom7 turnout7 union07 veto07 PR rgdppc07 fempar07 unempl07 , pcse, and

xtgls R07 p905007 stdiffmed7 turnout7 union07 veto07 PR rgdppc07 fempar07 unempl07 , pcse.

Note that this model adjusts for heteroscedasticity, but not for contemporaneous correlation of errors. The paper explains why this is optimal with our data. To estimate the same model in Stata8, use the following command:

xtgls R07 p905007 stcogcom7 turnout7 union07 veto07 PR rgdppc07 fempar07 unempl07, forcepcse.

To estimate the model with correction for contemporaneous correlation, change the command to:

xtpcse R07 p905007 stcogcom7 turnout7 union07 veto07 PR rgdppc07 fempar07 unempl07 , pairwise

Note that whenever a change is made to the variables in the model, it is necessary (unless you program) to change the dependent variable and all independent variables in tandem to find the value of Rho that maximizes the R-squared.

II. Partisanship:

There are two files, which can be used to replicate all the partisan results. These results are restricted to governments that are not scored as pure center governments according to the Castles-Mair classification (ie., getting a score of 3.0 = center in the C-M scheme). The first file contains additional data that is irrelevant to most uses – these can simply be ignored. Use the following information to replicate the results in the article:

Use "APSR06_1" to replicate tables 1 and 6 using the following State commands:

tab partysys cenleftg , row produces Table 1

tab partysys cenlftg2 , row produces Table 6

Use “APSR06_2" to replicate Table 7 using the following Stata commands:

collapse diffmeds cogcab partysys leftfrag Right_Overrepresentation turnout Union FEMPAR , by( CCODE)

Then run the regressions:

reg diffmed partysys if CCODE~=225 produces fist column of Table 7, etc.