3
$\begingroup$

I am writing my master´s thesis in finance on the topic of voluntary disclosure of financial targets in annual reports of manufacturing firms.

Context

  • I have created a dependent variable that is an index of level of disclosure.
  • At the moment I am gathering firm characteristics as independent variables from around 10 years (unbalanced panel data).

Questions

  • Should the model be probit of logit?
  • Should the model be fixed or random?
  • How should it be implemented in Stata?
$\endgroup$
  • 2
    $\begingroup$ A few questions: (1) "level of disclosure" and "index" imply a variable with more than two levels; have you considered using an ordinal scale as the dependent variable? (2) Does disclosure vary over time or is it operationalised as constant over time? $\endgroup$ – Jeromy Anglim Apr 29 '11 at 12:43
  • $\begingroup$ 1) The measurements of voluntary disclosure is indicated by a ratio from 0-1. 1 representing complete disclosure, 0 non. 2) Level of disclosure vary over time $\endgroup$ – BEF Apr 29 '11 at 23:13
  • $\begingroup$ Thus, your dependent variable is not "binary" and not "dummy", it is measured on a scale from zero to one with many values in between? $\endgroup$ – Jeromy Anglim Apr 30 '11 at 3:10
  • $\begingroup$ I have updated the question and question title to try to reflect the points you have made in the comments; feel free to edit your question, if I have misconstrued what you are saying. $\endgroup$ – Jeromy Anglim Apr 30 '11 at 3:27
  • $\begingroup$ @Jeromy Anglim - Yes, its true. $\endgroup$ – BEF May 1 '11 at 9:17
3
$\begingroup$

Just on the first of your questions: probit or logit?

In practice it usually makes little difference. You will get different parameter estimates from the two methods, but that is because the parameters mean different things. When you then use these to model, the differences will then largely disappear. The logistic density distribution is slightly leptokurtic (a sharper peak and fatter tails) compared with the normal distribution, and being comfortable with odds I personally find it slightly easier to explain logit methods, but you need not worry too much whichever you choose.

You can find some discussion in this lecture.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ (+1) A nice feature of the logistic versus probit is that it is relatively insensitive to a small number of responses that get incorrectly coded. This is, of course, related to the fact that it has fatter tails. $\endgroup$ – cardinal Apr 30 '11 at 22:49
  • $\begingroup$ Logistic is basically indistinguishable from a t-distribution with about 7-8 degrees of freedom. If you wanted "super-robustness" you should use the arctan link function - so you have a cauchy distribution $\endgroup$ – probabilityislogic May 1 '11 at 12:10
  • $\begingroup$ @probabilityislogic: How sure are you about your first statement? $\endgroup$ – cardinal May 1 '11 at 19:34
  • $\begingroup$ @cardinal - if you're referring to logistic and t-dist. here is a link $\endgroup$ – probabilityislogic May 2 '11 at 7:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.