Login  Register

Re: Second posting re: test of overall significance for multinomial LR

Posted by Ryan on Apr 22, 2011; 1:42am
URL: http://spssx-discussion.165.s1.nabble.com/Second-posting-re-test-of-overall-significance-for-multinomial-LR-tp4331823p4332193.html

The LOGISTIC procedure in SAS automatically provides a "Global Tests"
table, which includes a Likelihood Ratio Chi-Square test, a Score
Chi-Square test, and a Wald Chi-Square test.

Below I generate multinomial response data (3 categories) with a
single continuous predictor in SAS and then fit the model employing
the LOGISTIC procedure in SAS. If you run the code, you'll observe
that all three Chi-square statistics are provided. The Chi-Square
statistics are similar but not identical.

I decided to fit the same model using NOMREG in SPSS (code below) and
observed that the Likelihood Ratio Chi-Square Statistic was the same
as the one derived from SAS.

Ryan

--

data multinomial_response;
  beta1_0 = .2;
  beta1_1 = .5;
  beta2_0 = -.2;
  beta2_1 = .7;
  seed = 1234579;
   do id = 1 to 500;
        x = rannor(seed);
        eta1 = beta1_0 + beta1_1*x;
        eta2 = beta2_0 + beta2_1*x;
        prob1 = exp(eta1) / (1 + exp(eta1) + exp(eta2));
        prob2 = exp(eta2) / (1 + exp(eta1) + exp(eta2));
        prob3 =         1 / (1 + exp(eta1) + exp(eta2));
        call rantbl(seed, prob1, prob2, prob3, y);
        output;
      end;
  keep id x y;
run;

proc logistic data = multinomial_response;
class;
model y = x / link = glogit;
run;

-------
SPSS Code:

NOMREG y (BASE=LAST ORDER=ASCENDING) WITH x
  /MODEL
  /INTERCEPT=INCLUDE
  /PRINT=PARAMETER SUMMARY LRT.

On Thu, Apr 21, 2011 at 9:05 PM, Bruce Weaver <[hidden email]> wrote:

> As Ryan said, there will be a likelihood ratio test for the full model versus
> an intercept-only model.  You can tell which one it is by looking at the
> degrees of freedom, which will be equal to the number of model parameters,
> not including the intercept.  Judging by what you say, SAS gives a Wald test
> for the same model comparison (full vs intercept only).
>
> When I responded earlier, I understood you to be asking if the likelihood
> ratio test from SPSS was equivalent to the Wald test from SAS.  These two
> tests (and the Score test) are asymptotically equivalent, but in finite
> samples generally give somewhat different results.  So don't expect the
> likelihood ratio test from SPSS to match exactly the Wald test from SAS, if
> that is what you're asking.
>
>
>
> J McClure wrote:
>>
>> For SPSS it is a likelihood ratio test. My question is whether this is a
>> test of overall significance or whether there is a different measure in
>> SPSS that is equivalent to the SAS Type III Analysis of effects Wald
>> chi-square which I was taught is a test of overall significance for a
>> multinomial LR model.
>> Thanks!
>> Jan
>>
>> On 4/21/2011 3:00 PM, Bruce Weaver wrote:
>>> IIRC, the heading for that part of the output says "Likelihood Ratio
>>> Test",
>>> does it not?  If so, the test you're getting is on the change in -2LL,
>>> not a
>>> Wald test.
>>>
>>>
>>>
>>> J McClure wrote:
>>>> Hi,
>>>> I didn't get any responses to my first post. I hope I am using the
>>>> correct etiquette in re posting.
>>>> I'm running multinomial logistic regression using nomreg. Is the "model
>>>> fitting information" chi-square a test of overall significance in the
>>>> way that SAS Proc Logistic Type III Analysis of Effects with a Wald
>>>> chi-squared is an overall significance test.
>>>> Thanks for any help,
>>>> Jan
>>>>
>>>> =====================
>>>> To manage your subscription to SPSSX-L, send a message to
>>>> [hidden email] (not to SPSSX-L), with no body text except the
>>>> command. To leave the list, send the command
>>>> SIGNOFF SPSSX-L
>>>> For a list of commands to manage subscriptions, send the command
>>>> INFO REFCARD
>>>>
>>>
>>> -----
>>> --
>>> Bruce Weaver
>>> [hidden email]
>>> http://sites.google.com/a/lakeheadu.ca/bweaver/
>>>
>>> "When all else fails, RTFM."
>>>
>>> NOTE: My Hotmail account is not monitored regularly.
>>> To send me an e-mail, please use the address shown above.
>>>
>>> --
>>> View this message in context:
>>> http://spssx-discussion.1045642.n5.nabble.com/Second-posting-re-test-of-overall-significance-for-multinomial-LR-tp4331823p4331960.html
>>> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>>>
>>> =====================
>>> To manage your subscription to SPSSX-L, send a message to
>>> [hidden email] (not to SPSSX-L), with no body text except the
>>> command. To leave the list, send the command
>>> SIGNOFF SPSSX-L
>>> For a list of commands to manage subscriptions, send the command
>>> INFO REFCARD
>>>
>>
>> =====================
>> To manage your subscription to SPSSX-L, send a message to
>> [hidden email] (not to SPSSX-L), with no body text except the
>> command. To leave the list, send the command
>> SIGNOFF SPSSX-L
>> For a list of commands to manage subscriptions, send the command
>> INFO REFCARD
>>
>
>
> -----
> --
> Bruce Weaver
> [hidden email]
> http://sites.google.com/a/lakeheadu.ca/bweaver/
>
> "When all else fails, RTFM."
>
> NOTE: My Hotmail account is not monitored regularly.
> To send me an e-mail, please use the address shown above.
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Second-posting-re-test-of-overall-significance-for-multinomial-LR-tp4331823p4332172.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD