I'd like to estimate the factors that determine the development of heating networks in French municipalities. To do this, I use various variables (population density, age of dwelling, ideology of the municipality, heat consumption, fuel poverty, etc.) to determine the probability of the presence of a heating network in a municipality. I have all these variables for each year between 2001 and 2021, so I want to use the progit to estimate the different determinants. I have made the assumption here that the omitted variables are time-invariant but vary between municipalities, so I have chosen to run a fixed-effect regression. However, when I use pglm with within an error message appears:
probit_model
Error in maxRoutine(fn = logLik, grad = grad, hess = hess, start = start, :
the "start" argument is missing, with no default value.
However, the package documentation tells me that fixed-effect modelling exists with pglm
With within and between modelling the same message is displayed:
probit_model <- pglm(dh2000 ~ Foret + Densité + comgeother + com.proxincin + nouveaulog+ gauchemun + gauchedpt + gaucheepci + precanrj + proxdh + binaire_gauchereg + mundpt,
+ data = df_complet,
+ model = "within",
+ family = binomial(link = "logit"),
+ na.action = na.omit)
Erreur dans maxRoutine(fn = logLik, grad = grad, hess = hess, start = start, :
l'argument "start" est manquant, avec aucune valeur par défaut
Whereas with pooling and random as well as with the default value (I don't know which it is), I manage to get estimates. However, I think that given my problem, fixed-effect modelling is the most robust.
I then tried LSDV regression or the plm package, but they can't be used for binary dependent variables.
I'm looking for advice on how to run a fixed-effect logit regression with panel data on R for my study.