Announcement

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

  • Regression coefficient changes sign and becomes statistically significant when adding year and country fixed effect

    Dear Statalist!

    I have a dataset that covers loan applications from 2012 to 2017. The columns include information on both the loan (e.g., the interest rate, loan duration, loan amount) and the borrower (e.g., income, employment status, gender, age). Using a regression approach, I want to investigate whether gender has an effect on the interest rate of a loan.

    I have two different specifications of the regression model. The DV is the interest rate in both models. In the first model, I include a gender dummy (0 = male, 1 = female) and a set of loan and borrower-level controls. The second models uses the same gender and control variables as in the first model but here I add year and country fixed effects. For both models, I use the reghdfe command with vce(robust). In the first model, I use the noabsorb option and in the second model absorb(year country).

    When I look at the coefficient of the gender dummy in the first model, it is negative but not statistically significant. However, in the second specification where I include year and country fixed effects the coefficient flips sign and becomes statistically significant. I do not quite yet grasp how the inclusion of the fixed effects contribute to the flipping of the sign and the coefficient becoming statistically significant. What could be an explanation? Or what further analyses or testing could I do to figure this out?

    Many thanks!
    Last edited by Timo van Amen; 19 Aug 2023, 07:07.

  • #2
    Cross-posted and already answered at https://stats.stackexchange.com/ques...stically-signi

    Please note our request in the FAQ Advice that you tell us about cross-posting. This works two ways: people interested in answering -- and people interested in an answer -- can see what has already been said and benefit -- either by not posting something already said as well or better, or by finding out about previous answers.
    Last edited by Nick Cox; 19 Aug 2023, 08:28.

    Comment


    • #3
      Timo:
      welcome to this forum.
      I can just echo Nick's reminder about reading the FAQ and act on them accordingly, but on a different note: please post what you typed and what Stata gave you back. Thanks.
      Please note that we can know your data and related problems conditional on your description (and, unfortunately, words are not that helpful when it comes to quantitative analysis).
      Kind regards,
      Carlo
      (Stata 19.0)

      Comment

      Working...
      X