Differences in result with linktest and Ramsey reset test

[Laiy Kho]

Hello,

I employed an OLS regression on stata with cross sectional data. I ran the linktest and Ramsey's reset test (using estat ovtest command). However, I obtained different results. Linktest suggests that there are no specification errors as the p-value is insignificant. However, Ramsey's reset test is rejected.

linktest

Source | SS df MS Number of obs = 954,094
-------------+---------------------------------- F(2, 954091) > 99999.00
Model | 260.242349 2 130.121174 Prob > F = 0.0000
Residual | 251.502479 954,091 .000263604 R-squared = 0.5085
-------------+---------------------------------- Adj R-squared = 0.5085
Total | 511.744828 954,093 .000536368 Root MSE = .01624

------------------------------------------------------------------------------
Y | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
_hat | .9907048 .0080616 122.89 0.000 .9749042 1.006505
_hatsq | .0585772 .0504061 1.16 0.245 -.040217 .1573715
_cons | .0003513 .0003144 1.12 0.264 -.000265 .0009675
------------------------------------------------------------------------------

. estat ovtest

Ramsey RESET test for omitted variables
Omitted: Powers of fitted values of Y

H0: Model has no omitted variables

F(3, 954083) = 66829.11
Prob > F = 0.0000

What does the differences in test results imply? Are the assumptions of OLS violated? Which test result should I give preference to?

Thank you

[Carlo Lazzaro]

Laiy:
-estat ovtest- includes the normalized fitted values squared, cubic and ^4 (the last two powers are of limited utility in real world research). In addition, unlike -linktest-, -estat ovtest- cannot take heteroskedasticity into account;
-linktest- consider the linear and the squared fitted values (with no normalization).
Therefore, different outcomes are possible.
That said, I would trust the -linktest- outcome.