Announcement

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

  • Interpreting a regressors` direct and indirect effect when using an endogenous covariate

    Dear experts,

    currently I am working on a project in which we evaluate firm-specific resources towards their effect on firm performance. Therefore, we regress a set of covariates (education, experience, etc.) against the business’s growth rate of employees. In other words, we would like to find variables which are positively linked to a firms employment growth. We do so by using simple OLS.
    In addition, we include a binary dummy variable indicating whether the firm received funding from a venture firm. Venture firms are believed to not only invest money, but also expertise. Yet, we believe this venture financing is endogenous to firm size due to unobserved firm heterogeneity and/or reverse causality. Thus, simply using OLS will lead to biased estimates when we include the VC-dummy into our regression.

    My ultimate research outcome should be an answer to the question: What direct effect for example education has on the firms employment growth rate an to what extent (indirect effect) the venture financing has in case the firm was able to acquire such.

    Can you recommend a concept or model that I can apply in Stata in order to “evaluate” the change in effect-size for example education has on the firm`s employment growth once I include the endogenous venture capital dummy? My current database consists of only 90 sample firms for which I have employment data for only the time of the survey as well as the time the firm was founded. Thus, I do not have a real panel dataset.
    I would much appreciate your thoughts and/or ideas!
    Thanks in advance,
    Alex

  • #2
    You need instrumental variable analysis which can be acomplised using Stata's -ivregress-, -ivreg2- or Structural Equation Modeling (-sem-) commands. Read the related help files i.e search for help sem . Personaly I prefer -sem- as it provides lot more flexibilities. The test of direct and indirect effects are very simple to estimate. But I will read the help file first as there are key concepts that need to be understood. In addition to the help file, a good book to start for -sem- will be, Discovering Structural Equation Modeling Using Stata, Alan C. Acock, 2013
    Roman

    Comment


    • #3
      To add a couple of observations. With a dichotomous endogenous variable, some of the normal routines may be problematic - you need a routine that explicitly handles such dichotomous endogenous variables. SEM or GSEM will handle these. You may have a real problem with the sample size. Some (most?) of the relevant techniques only have asymptotic properties. You also need to be careful you don't have too many variables relative to the sample size.

      Comment

      Working...
      X