1

I am doing a Poisson regression analysis and have found that the type of programme a student is enrolled on to does not have a significant effect on the outcome. When it comes to writing out the fitted equation would I need to include the programme if it does not have an effect?

chl
  • 50,972
  • 18
  • 205
  • 364
Emily
  • 151
  • 1
  • 1
  • 3
  • 5
    An interesting change occurs between the beginning and end of your question, Emily. Initially you refer to a "[not] significant effect." That means you could not *estimate* the effect with *sufficient reliability* given the data you have. At the end, though, you appear to use "effect" in the sense of the *true* effect, not its measurement. The two differ in an important sense: the former is your estimate, the latter is real (but unknown). If we now reread your question, understanding the last "effect" in the *first* sense, it may clarify the situation. – whuber Feb 08 '13 at 16:54

1 Answers1

2

When you report on your statistical analysis, yes, you should report the model you based your conclusions on, not drop components that are not statistically significant - those may still influence the effect of other regressors.

If you are really asking whether you should drop the insignificant predictor and re-fit your model (called "stepwise regression") - you should not do that if your goal is inference, but this may be defensible if your goal is prediction.

Community
  • 1
Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357