 Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: Box-Tidwell Test

 From Nick Cox To statalist@hsphsun2.harvard.edu Subject Re: st: Box-Tidwell Test Date Tue, 26 Apr 2011 10:03:32 +0100

```Thanks for this, but still no specification of commands used, nor of
what you typed.

My point remains: there is no sense in which any one-to-one
transformations of 0-1 variables change their key statistical
properties. The predictors' coefficients would change, but the
statistical merit of the resulting model remains the same. Otherwise
put, your concern is to test linearity, but linearity properties with
0-1 predictor variables are invariant under one-to-one transformations
for the reasons I gave. Thus I can't see that there is a statistical

Here's a dopey example. (You can confirm for yourself that the same
applies with two dummies.)

. sysuse auto
(1978 Automobile Data)

. su weight

Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
weight |        74    3019.459    777.1936       1760       4840

. gen hiweight = weight > r(mean)

. logit foreign hiweight

Iteration 0:   log likelihood =  -45.03321
Iteration 1:   log likelihood = -32.869402
Iteration 2:   log likelihood = -31.855709
Iteration 3:   log likelihood =  -31.79108
Iteration 4:   log likelihood = -31.790431

Logistic regression                               Number of obs   =         74
LR chi2(1)      =      26.49
Prob > chi2     =     0.0000
Log likelihood = -31.790431                       Pseudo R2       =     0.2941

------------------------------------------------------------------------------
foreign |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
hiweight |  -3.205453   .8022163    -4.00   0.000    -4.777768   -1.633138
_cons |   .2876821    .341565     0.84   0.400    -.3817731    .9571372
------------------------------------------------------------------------------

. replace hiweight = cond(hiweight == 1, -999, 42)

. logit foreign hiweight

Iteration 0:   log likelihood =  -45.03321
Iteration 1:   log likelihood = -32.869402
Iteration 2:   log likelihood = -31.855709
Iteration 3:   log likelihood =  -31.79108
Iteration 4:   log likelihood = -31.790431

Logistic regression                               Number of obs   =         74
LR chi2(1)      =      26.49
Prob > chi2     =     0.0000
Log likelihood = -31.790431                       Pseudo R2       =     0.2941

------------------------------------------------------------------------------
foreign |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
hiweight |   .0030792   .0007706     4.00   0.000     .0015688    .0045896
_cons |   .1583554     .32909     0.48   0.630     -.486649    .8033599
------------------------------------------------------------------------------

On Tue, Apr 26, 2011 at 9:40 AM,
<lisenger@students.mail.uni-mannheim.de> wrote:
> Hi
>
>
> Sorry for the vague explanations.
>
> I want to check in how far being interested in a political scnadal and
> evaluating the scandal as an important problem, impacts the vote intentions
> of the people. My exampekl is the CDU- party finance scandal in Germany in
> 1999/2000.
>
> My dependent variable is vote intention
>
> 1 = vote intention for CDU/CSU 0 = vote intention for other parties
>
> Independent Variables:
>
> subject orientation: 1 = interested in scandal 0 = not interested
> problem orientation: 1 = scandal importatnt problem 0 = scandal not
> important
>
>
> My logistic regression: logit voteUnion subject problem
>
> My "not working" I mean that the Box-Tidwell test does not give any results.
> Normally below the regression, there would be columns with the results, yet
> in my case nothing
>
>
> Iteration 0:  Deviance =  46385.51
> Iteration 1:  Deviance =  46385.51 (change =         0)
>
> [Total iterations: 0]
>
> Box-Tidwell regression model
>
> Logistic regression                               Number of obs   =
>  36278
>                                                  LR chi2(2)      =
> 228.82
>                                                  Prob > chi2     =
> 0.0000
> Log likelihood = -23192.755                       Pseudo R2       =
> 0.0049
>
> ------------------------------------------------------------------------------
>   wabsunion |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
> Interval]
> -------------+----------------------------------------------------------------
>      themen |  -.3546749   .0278373   -12.74   0.000     -.409235
> -.3001148
>     problem |  -.3162207   .0492572    -6.42   0.000    -.4127631
> -.2196784
>       _cons |  -.5600199   .0126023   -44.44   0.000    -.5847199
> -.5353199
> ------------------------------------------------------------------------------
> Deviance:46385.509.
>
> .
> Here are the results.
>