Announcement

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

  • Ommited variables fixed effects regression

    Hello all,

    I am a beginner in stata and using it for my bachelor thesis. Therefore my questions are rather basic. I hope I will still find help in this forum.

    I use a regression with two fixed effects (year and company) and my data is distributed over time and firms. Now I was asked to add further independent variables and loss and big were suggested. Both are binary and indicate whether the respective firm is big (loss making). loss2 and big2 are loss(big)*timedifference:

    reghdfe ln_deviation ln_timedifference loss loss2 big big2, absorb (cusip yearforecast)

    Stata is omitting those variables due to collinearity. I have found the explanation online that this might be the case as the deviation of those variables over time are rather limited. Does this explanation make sense? And how can I adjust for this (exclude the variable / change the nature of this variable / do nothing)?

    Furthermore, I use the two natural logarithms to make the interpretation easier. This reduces my standard error and makes some of my analysis insignificant. Is there a rule of thumb to test whether the natural logarithm fits the distribution of my data?

    I am very thankful for any kind of help.

  • #2
    You didn't get a quick response. You'll increase your chances of a helpful response by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

    With just year and cusip, you can use xtreg and include i.year among the regressors. If a variable doesn't vary within panels, it is colinear with the fixed effects and cannot be estimated - the fixed effects control for everything that varies by panel but not within panel.

    Taking logs means you are running a different model. Whether taking logs makes it easier to understand is up to you. My personal preference is for the original variables. The issue is not whether logs fit the data or not. It is whether the relation being estimated follows the transformations you've made on the variables.

    Comment

    Working...
    X