Linktest

[Benjamin Krüger]

Hey guys,

I have a question about the output of the linktest. I know that a significant _hatsq in combination with an insignificant _hat indicates a missspecification of the model but in my case I got significant results for the _hat (5%-level) and the _hatsq (1%-level). What does that mean? Do I have to adjust my model because the p>|z| is smaller for the _hatsq?

Thanks in advance
Ben

[Carlo Lazzaro]

Benjamin.
as recommended by the FAQ posting what you tyoed and what Stata gave you back would enormously help interested listers to reply positively.
That said:

test _hatsq reaches statistical significance, you actually have a misspecification issue

[Benjamin Krüger]

Hey Carlo,

thank you for your answer. I use a binaray logistic regression with robust standard errors:

 logit D_Sparbuch Sex Birthyear Upperdegree,vce(r)
Linktest 

Logistic regression Number of obs = 1,457
LR chi2(2) = 21.26
Prob > chi2 = 0.0000
Log likelihood = -812.78606 Pseudo R2 = 0.0129

------------------------------------------------------------------------------
D_Sparbuch | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_hat | -5.853248 2.541893 -2.30 0.021 -10.83527 -.8712305
_hatsq | 3.16496 1.170371 2.70 0.007 .8710752 5.458845
_cons | 3.563412 1.348668 2.64 0.008 .9200706 6.206754
------------------------------------------------------------------------------

[Carlo Lazzaro]

Benjamin:
did you try:

logit D_Sparbuch Sex c.Birthyear##c.Birthyear Upperdegree,vce(r)

[Benjamin Krüger]

Hey Carlo, thank you so much. It worked for this example

I want to add a question about the linktest. What happens if both _hat & _hatsq are insignificant? In my online search I found two examples which indicate that the linktest is only important for the _hatsq (both passed the linktest with an insignificant _hat) but on another site I found that an insignificant _hat indicates a missspecification.

My regression:
Code:
logit D_Betriebsvermögen RBmean sex if syear == 2007 & Entscheider_final == 1 ,vce(r)
My output:



HTML Code:
D_Betriebsvermögen Coef. Std. Err. z P>z [95% Conf. Interval] _hat -.2265207 .658621 -0.34 0.731 -1.517394 1.064353 _hatsq -.2748453 .1457805 -1.89 0.059 -.5605699 .0108793 _cons -1.197294 .703436 -1.70 0.089 -2.576004 .1814148

thanks in advance

Ben

[Benjamin Krüger]

[QUOTE=Benjamin Krüger;n1593421]Hey Carlo, thank you so much. It worked for this example

I want to add a question about the linktest. What happens if both _hat & _hatsq are insignificant? In my online search I found two examples which indicate that the linktest is only important for the _hatsq (both passed the linktest with an insignificant _hat) but on another site I found that an insignificant _hat indicates a missspecification.

My regression:

logit D_Betriebsvermögen RBmean sex if syear == 2007 & Entscheider_final == 1 ,vce(r)

My output:

D_Betriebsvermögen    Coef.    Std. Err.    z    P>z    [95% Conf.    Interval]
                        
_hat    -.2265207    .658621    -0.34    0.731    -1.517394    1.064353
_hatsq    -.2748453    .1457805    -1.89    0.059    -.5605699    .0108793
_cons    -1.197294    .703436    -1.70    0.089    -2.576004    .1814148

thanks in advance

Ben

[Carlo Lazzaro]

Ben:
the Methods and formula section of -linktest- entry in Stata .pdf model focuses on the statistical significance of sq_hat only.
Your results do not show evidence of model misspecification.

[Benjamin Krüger]

Thanks again Carlo

Your answers were really helpful