Poisson vs. linear regression

[lisa Uuuz]

Hello everyone,

I'm a newcomer to the forum and relatively inexperienced with Stata.
Currently I am writing my Master thesis and I´m a bit confused about which statistical method to use.
My research involves a dependent variable, "Number of employees participating in further training," and my main independent variable, "Number of vacant positions." Initially, I began my analysis with linear regression, assuming it would be suitable for a continuous dependent variable.
However, after reading an article that suggests using Poisson regression or Negative Binomial models for continuous discrete data, I'm contemplating whether linear regression is still appropriate or if the results might be inaccurate.

If the consensus leans toward Poisson regression, I'm curious to know if the implementation is similar to linear regression, or if there are specific nuances I should be aware of.
Any insights or guidance on the most fitting statistical approach for my scenario would be highly appreciated.

Thank you in advance for your assistance!

[Jeff Wooldridge]

Please provide more information on your outcome (dependent) variable. Do you see lots of zeros?

If there are zeros, you can apply a linear model directly to y and also use Poisson regression for y (which uses an exponential model). The coefficients have different meanings, but you can turn the Poisson estimates into average partial effects and compare them with linear. I think the exponential mean is more appealing when y >= 0. The coefficients have percentage interpretations, fitted values are never negative.

Use robust standard errors in all cases.

[lisa Uuuz]

Hello Jeff,

Thank you very much for your answer!
Yes, I see a lot of zeros (the 10th percentile is at 0). Am I correct in understanding that I can still proceed with linear regression and am not necessarily required to opt for a Poisson regression?
Or is the Poisson regression as an exponential model more appropriate, and therefore I need to use it?