Announcement

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

  • Problem choosing regression model for panel data with two main variables of interest and a lot of zeros (around 50/80) in one of them

    Hi all,

    I would like to ask you for your help on the following problem. I am conducting research on the adoption of electric vehicles. For this I have collected data for 8 years (2010 up till 2017) from 10 EU countries (Austria, Belgium, Denmark, France, Germany, Italy, the Netherlands, Sweden, Norway and the UK). The main variables of interest in this research are subsidy that governments give to incentivize consumers to buy an EV, or tax cuts/exemptions that are put in place for this purpose. Those two variables I have constructed apart from each other, because I want to find out if there is a difference in effectiveness of those policy measures to bring people to buy EV's instead of internal combustion engine vehicles.
    My dependent variable is the amount of newly registered electric cars for each year per country.
    At first I found out that the difference in effectiveness of the two kinds of policies could be found by performing a multiple regression analysis. Then I looked at my data and saw that it is does not fit the assumption of homoskedasticity, but that I have a lot of zeros in the Subsidy-variable. Because of that a (zero-inflated) Poisson Regression or Negative Binomial Regression would be the better fit to analyse my data, but is it possible to analyse the relative effectiveness of both variables of interest with those two or with one of those sorts of regression?

    Just for your interest I would like to mention that I use Stata 14.1.

    Thanks in advance for your time and answers!

    Kind regards,
    Henri
    Last edited by Henri Slob; 20 Jun 2018, 03:10.

  • #2
    Welcome to the Stata Forum / Statalist,

    Since you have panel data, it seems - xtpoisson - or - xtbnreg - models should come to the foreground.

    Please read the FAQ. There you’ll find information about sharing data/command/output.

    I assumed there are counts of yvar by year by state.
    Best regards,

    Marcos

    Comment


    • #3
      Hello Marcos,

      Thank you for affirming my idea about using a Poisson or Binomial regression. As you stated correctly I have counts by year by state, so 8 years times 10 countries which adds up to 80 datapoints per variable.
      With this I think I will be able to conduct my research and get valid results.

      Kind regards,
      Henri

      Comment

      Working...
      X