Insignificant results

[Anuradha Saikia]

Hello everyone

For a study of mine, my main independent variable is showing as insignificant ( the main relationship I am studying) in fixed effects , pooled and random effects in panel regression. Is there a way to try other models to make it significant.

[Carlo Lazzaro]

Anuradha:
short answer: no.
A more detailed reply would recommend to:
1) share what you tyoed nad what Stata gave you back (as per FAQ);
2) check whether your model is corerctly specified, so to give a fair and true view of the data generating process.

[Anuradha Saikia]

this is something I got .

[Anuradha Saikia]

also I got this Carlo Lazzaro

[Anuradha Saikia]

I do not want to drop this hypothesis which is important for my study. very worried now

[Carlo Lazzaro]

Anuradha:
1) despite the evidence of a panel-wise effect, your within-Rsq is low;
2) predictors are expressed in different metrics, and this may contribute to explain the weird results;
3) I would test the chance of turning point for -firmage- and -firmsize- (linear and square terms);
4) with 148 panels, cluster-robust standard errors are recommended.

[Anuradha Saikia]

This means I check the predictors formulae once. For eg ROA is in ratio then multiplied by 100, . contingent liabilities is taken as it is .i don't know if taking log of it is the right approach. i will check for robust standard errors .

[Carlo Lazzaro]

Anuradha:
my last relationships with corporate finance date back to 1990: hence, I cannot say whether the literature in your research field consider different metrics in the same regression as absolutely legal.
A log-linear model may make sense as it expresses the variation of the regressand in %.
Agan, I would search for possible turning points and apply cluster-robust standard errors.

[Anuradha Saikia]

Thank you for the help Carlo Lazzaro . i will check up all possibilities

[Anuradha Saikia]

Carlo Lazzaro if you could look the results this time. i habe taken robust standard errors and changed the variable formulae.

[Anuradha Saikia]

whereas If I dont cluster standard errors I get is this.

[Carlo Lazzaro]

Anuradha:
I'd feel more comfortable with your code in #10 (with cluster-robust standard errors).
In addition:
1) is -FirmAge- a time-invariant precitor?
2) Did you test as a predictor:

c.Firmsize##c.Firmsize

[Anuradha Saikia]

Firm Age is time invariant (age of each firm in 2022). yes I included it as control.for cluster robust standard errors is a p=0.194 for CL (which is contingent liabilities) is coming as insignificant. i do not know why .

[Carlo Lazzaro]

Anuradha:
you can get rid of -FirmAge- then.

[Anuradha Saikia]

Okay Carlo Lazzaro thank you again.