Failed ovtest but insignificant hatsq in linktest

[Medel Valiente]

Sirs: I have a different result from ovtest and linktest. What should i do? Sorry, i am new to Stata. Thank you for your answers.

[Carlo Lazzaro]

Medel:
both newbies and experienced Stata users are recommended by FAQ to post what they typed and what Stata gave them back (via CODE delimiters) to increase their chances of getting helpful replies.
The same source warns about attachments. Thanks.

[Medel Valiente]

Sir Carlo:
Sorry. How about these images?

[Carlo Lazzaro]

Medel:
whereas -_hatsq- has no explanatory power, there may still be an omitted linear predctor in your regression equation.
This is the only relevant issue, in that both -hettest- and -vif- (even though for good reasons - see https://www.amazon.com/Course-Econom.../dp/0674175441 - this latter post regression test has recently lost a bit of its popularity) look OK.
Two asides to what above:
- for the future, please use CODE delimiters instead that resulst screenshots to explain what's the matter with your data;
- please call me Carlo, just like all on (and many more off) the list do.

[Medel Valiente]

Carlo:
Thanks for the tips. I will learn how to use CODE delimiters.
Another question, if i remove one variable, the p-value from ovtest becomes insignificant. Is it because of collinearity?

The things is, i have two options.
1) remove a variable, p-value is insignificant
2) fail ovtest, but continue to explain that my model needs additional variables to increase its explanatory value

Which is better? Is the latter a pure violation?
Many thanks!

[Carlo Lazzaro]

Medel:
the predictors included in right-hand side of the regression equation sholuld give a fair and true view of the underlying data generating process: hence, loking for the "right p-value" (whatever that may mean) it is ok conditional on including all the predictors that are considered relevant in your research field, which remains te main issue.

[Marcos Almeida]

Carlo gave a very insightful reply.

On a slightly different note, I'm wondering about the fact that the model, in spite of just 60 observations for 6 predictors, presents significant coefficients for half of the predictors. Add to it that the R-squared is impressively high, all this for a model that, potentially, would "benefit" if adding (theoretically) ommited variables. I don't know what each predictor represents, and surely the field is not the one I'm more acquainted with, but, well, I'm wondering whether this is not sort of "too good to be true" situation.