Interpretation of logit model

[Jordi Imbrechts]

Dear all,

For my thesis I have to research the motives of a firm to execute a layoff (>= 10% of workforce being dismissed). Data consists of panel data for a 10 year period of firms.
When I ran regress, I obtained the following results:

Of course I know that for the final interpretation, I will have to look at the oddsratio. But before doing this, I wanted to look if there is a positive or negative effect for my independent variables on layoff...
- With variables such as Lnleeft (=LnAGE in English), it is quite easy: the younger the firm, the more chance of a layoff.
- But my main problem arises when I want to interpret the sign of "dalingProd2J" (= decrease in productivity for 2 consequent years in English). Does the negative sign here mean that when a firm has a decrease in productivity for 2 following years, the chance of a layoff will be bigger?

The combination of "decrease" and "-" confuses me a little bit here...

Kind regards,
Jordi

[Tom Scott]

Please show a summary table for the variables in your model so we know how everything is measured. Was dalingProd2J originally a categorical variable along with dalingProd3J and maybe dalingProd1J?

[Jordi Imbrechts]

Dear Tom, thanks for the response.

dalingPROD2J & dalingPROD3J are both dummy variables that indicate if a firm has a decrease in productivity for 2 consequent years (and also for 3 years). = 1 when there is a decrease for 2 (or 3) consequent years.
This dummy is generated based on the continous variable "OmzetWN_win1" (=Productivity in English).

Thanks in advance

[Carlo Lazzaro]

Jordi:
as an aside to Tom's helpful advice, I woul consider two other issues:
1) if you have panel data, why going -logit- instead of -xtlogit-?
2) even assuming that (pooled) cross-sectional logistic regression is the way to go due to the lack of evidence of a panel-wise effect, why not clustering your standard errors on -panelid-, as yhe observations belonging to the same panel are not independent?
As a sidelight, using odds-ratio instead of coefficients makes the interpretation of your results easier.

[Jordi Imbrechts]

Dear Carlo, thanks for your response.

I am using xtlogit, it could be perhaps that I uploaded the wrong screenshot (when I was trying it with just -logit-). My bad.
Thanks for the advice.

[Tom Scott]

So to answer your question, you found that firms experiencing two years of decreased productivity are less likely to experience a lay off compared to those not experiencing two years of decreased productivity, holding all other covariates constant. It doesn't look like your variable is lagged, though, so can you be sure the firm experienced a full 2 years of decreased productivity prior to the layoff?

[Jordi Imbrechts]

Thank you for the comment Tom!