61
dependent variable equals "1" if the province of the sponsor is mentioned and "0"
otherwise.
My principal covariates lie at different levels of aggregation. For the municipal
model, the first set, including the main predictor, belongs to the legislator-level:
background as a mayor, pertinence to the majority party, pertinence to the governor's
party in her province, absolute distance from the median ideal point in the floor,
member of the Peronist party and member of a Provincial party. At the province-level, I
include a main covariate that differentiates districts well: district magnitude for Federal
Deputies. Since variation across provinces is huge in many aspects (institutions, political
patterns of continuity, turnout), and a specification including every single factor would
deviate the analysis from the main focus of this study; I prefer to let the estimations vary
by province. Thus, I run a random intercept multilevel model that captures the
province-specific effects over the likelihood of progressively ambitious politicians
submitting targeted legislation. Given that the structure of the dependent variable is
binary, I will use a Bernoulli- logistic function in the systematic component of my
equations. As an alternative estimation, I use an ordinary logit model with legislator-
level clustered standard errors, in order to take individual-level variation into account.
The clusters are made at the legislator-legislative period level (four years)36. The
employment of these grouped errors lets me capture the Iegisla tor-to-legisla tor
differences. Since I do not think that time makes any difference here (I am not theorizing
any kind of learning process or a dependence of current values of past legislative
activity), I am not including any time-series parameters.
361 thank Brian Crisp for pointing out the importance of running a model like this.