Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • PPMLHDFE - Perfect Multicollinearity with Fixed Effects

    Dear All,

    I am currently working on my dissertation about the effect of environmental policy stringency index on Turkey`s inward FDI stocks. For the study, I employed panel data of OECD and BRIICS countries from 2009-2020. I utilised PPML for zero values in my dataset, and I am using the code ppmlhdfe in Stata. I am new to this estimation technique and having some problems. The problem that I am encountering is that my fixed effects are perfectly multicollinear with some of my variables. For instance, when I employ time-fixed effects, Turkey`s GDP, POP, HDI, infrastructure, inflation and institutional quality variables are omitted from the model. In addition, when I control for country-fixed effects, the religion proximity index and distance variables are omitted. I am trying to find a way to both control fixed effects and have the relevant variables in my model.

    1) Can some things be improved about how I constructed the data? Is it normal to encounter such a problem?

    2) One of my ideas was to construct variables in relative terms for GDP, POP, HDI, infrastructure, inflation and institutional quality. Is it a good strategy? What can be done for religion proximity and distance variables? I would appreciate it if you could share your ideas with me.

    Also, unfortunately, my sample size is small.

    Dear Joao Santos Silva, I would appreciate it a lot if you could help me with this.

  • #2
    Dear Merve Ozcam,

    What you describe is a very normal situation: when we include different sets of "fixed effects" some of the variables of the model drop out because they are collinear with the fixed effects. This is not a problem, unless you are particularly interested in the coefficients of the variables that drop out. The problem will be particular severe when the data has a single destination or a single origin, which appears to be your case. You should think carefully about the model you want to estimate and the parameters you want to identify, and then find suitable data.

    Best wishes,

    Joao

    Comment

    Working...
    X