Advice regarding Ramsey RESET test result

[Guest]

Hello,

I'm running an OLS regression with around 2000 observations and a few variables - most of them are categorical variables.

When the dependent variable is regressed against certain independent variables (e.g. variables A, B, C and D), the model passes the RESET test.

However, when an additional variable is added (e.g. variable E), it fails the test. This additional variable happens to be significant and increase the adjusted R-squared significantly.

Why would this be the case? Could it be because the new variable, although significant, is non-linear?

Thank you,

Guest

[Carlo Lazzaro]

Guest:
welcome to this forum.
If the variable E is continuous, you can investigate whether -estat ovtest- does not reach statistical significance when you add a squared term for E:

regress <depvar> <other predictors> c.E##c.E
estat ovtest

[Guest]

Hello Carlo,

Thank you very much for your reply.

I've entered the following command (hopefully this is correct):

regress log_price_realised_usd log_area_size signed dated monogram inscribed ib(freq).sup2 ib(freq).att c.lot_notes_word_count##c.lot_notes_word_count

The regression output is that <lot_notes_word_count> and <c.lot_notes_word_count#c.lot_notes_word_count> are both significant (p = 0.000).

The RESET test result is:

Ramsey RESET test using powers of the fitted values of log_price_realised_usd
Ho: model has no omitted variables
F(3, 1927) = 22.63
Prob > F = 0.0000

What can I conclude from this result? Is the RESET test telling me that, because the squared term of variable E is significant, then variable E can't be linear? Does this mean that I can't use this variable in my model? The variable may not be linear, but it's still significant, so one should be able to draw some conclusions from its significance, no?

Thank you,

Guest

[Carlo Lazzaro]

Guest:
the result of -estat ovtest- is telling you that the model is misspecified.
However, it may well be that the including both the linear and the squared term for -E- variable makes sense.
My advice is to check that your set of predictors is consistent with the data generating process that you're investigating (see what other researchers did in the past when presented with the same research topic).
It may also be that logging the regressand causes the -estat ovtest- outcome to reach statistical significance.
As an aside, for the future please share what you tyoed and what Stata gave you back via:

CODE delimiters (see the FAQ)

Thanks.