Extension of McNemar test for multiple matching

[Margaret MacDougall]

  Hello

  I would like to use an extension of McNemar's test for multiple matched controls (as specified in Pike and Morrow 1970). I am not a SAS user. Can anyone kindly advise me on the availability of procedures within SPSS for carrying out the above test (and indeed for calculating appropriates odds ratios and CIs). Failing this, if you know of other packages with these capabilities, I would be happy to hear from you. Of course, I could write my own macros but it would be good if I didn't have to reinvent the wheel.

  Best wishes

  Margaret




---------------------------------
 Yahoo! Answers - Get better answers from someone who knows. Tryit now.

[Marta Garcia-Granero]

Hi Margaret:
>   I would like to use an extension of McNemar's test for multiple matched controls (as specified in Pike and Morrow 1970). I am not a SAS user. Can anyone kindly advise me on the availability of procedures within SPSS for carrying out the above test (and indeed for calculating appropriates odds ratios and CIs).
If you have one case & multiple controls (the number of controls can
vary, but the number of cases per matching stratum must be one) you can
use conditional logistic regression (see Hosmer&Lemeshow book on
logistic regression models for technical details). It's easy with SPSS.
Raynald added to his page a simple example I posted to this list years ago:

http://www.spsstools.net/Syntax/RegressionRepeatedMeasure/ConditionalLogisticRegression.txt

If the number of cases varies you can also use Mantel-Haenszel OR
estimator. Ask for more details (sample dataset and SPSS code) if you
are interested in this last approach (I don't have it right now at hand,
that's why don't provide it; be patient, I'm kinda busy right now and
will take some hours to answer, sorry).

HTH,
Marta

[Margaret MacDougall]

Dear Marta

  Many thanks for your message. I am happy with conditional logistic regression but would be very grateful to receive the sample dataset and code you mention. I should mention that it is the number of controls per case that I am assuming would be varying.

  Best wishes

  Margaret

Marta Garcia-Granero <[hidden email]> wrote:
  Hi Margaret:
> I would like to use an extension of McNemar's test for multiple matched controls (as specified in Pike and Morrow 1970). I am not a SAS user. Can anyone kindly advise me on the availability of procedures within SPSS for carrying out the above test (and indeed for calculating appropriates odds ratios and CIs).
If you have one case & multiple controls (the number of controls can
vary, but the number of cases per matching stratum must be one) you can
use conditional logistic regression (see Hosmer&Lemeshow book on
logistic regression models for technical details). It's easy with SPSS.
Raynald added to his page a simple example I posted to this list years ago:

http://www.spsstools.net/Syntax/RegressionRepeatedMeasure/ConditionalLogisticRegression.txt

If the number of cases varies you can also use Mantel-Haenszel OR
estimator. Ask for more details (sample dataset and SPSS code) if you
are interested in this last approach (I don't have it right now at hand,
that's why don't provide it; be patient, I'm kinda busy right now and
will take some hours to answer, sorry).

HTH,
Marta



---------------------------------
Which email service gives you unlimited storage?

[Marta Garcia-Granero]

Hi Margaret:
> I am happy with conditional logistic regression but would be very
> grateful to receive the sample dataset and code you mention. I should
> mention that it is the number of controls per case that I am assuming
> would be varying.

* We start from the same dataset used for C. logistic regression *.

DATA LIST FREE /pair(f4.0) exposici (f4.0) outcome (f4.0).
BEGIN DATA
 1 0 0 1 1 1 2 0 0 2 1 1 3 0 0 3 1 1 4 0 0 4 1 1  5 0 0  5 0 1
 6 1 0 6 0 1 7 1 0 7 1 1 8 1 0 8 1 1 9 1 0 9 1 1 10 1 0 10 1 1
END DATA.
VALUE LABELS exposici
 0 "No"
 1 "Yes".
VALUE LABELS outcome
 0 "Control"
 1 "Case".

* Look at the last table and ignore the rest *.
CROSSTABS
  /TABLES=exposici  BY outcome  BY pair
  /FORMAT=NOTABLES
  /STATISTIC=CMH(1).

* If we compare the output with the one obtained with conditional
logistic regression *.
COMPUTE ftime=1+(outcome=0).
EXECUTE.
COXREG  ftime  /STATUS=outcome(1)
  /STRATA=pair
  /METHOD=ENTER exposici
  /PRINT=CI(95).

* You can see that the odds-ratio and its 95%CI is the same with both
analyses .

Best regards,
Marta

[Margaret MacDougall]

Many thaks for the info, Marta.

  Best wishes

  Margaret

Marta Garcia-Granero <[hidden email]> wrote:
  Hi Margaret:
> I am happy with conditional logistic regression but would be very
> grateful to receive the sample dataset and code you mention. I should
> mention that it is the number of controls per case that I am assuming
> would be varying.

* We start from the same dataset used for C. logistic regression *.

DATA LIST FREE /pair(f4.0) exposici (f4.0) outcome (f4.0).
BEGIN DATA
1 0 0 1 1 1 2 0 0 2 1 1 3 0 0 3 1 1 4 0 0 4 1 1 5 0 0 5 0 1
6 1 0 6 0 1 7 1 0 7 1 1 8 1 0 8 1 1 9 1 0 9 1 1 10 1 0 10 1 1
END DATA.
VALUE LABELS exposici
0 "No"
1 "Yes".
VALUE LABELS outcome
0 "Control"
1 "Case".

* Look at the last table and ignore the rest *.
CROSSTABS
/TABLES=exposici BY outcome BY pair
/FORMAT=NOTABLES
/STATISTIC=CMH(1).

* If we compare the output with the one obtained with conditional
logistic regression *.
COMPUTE ftime=1+(outcome=0).
EXECUTE.
COXREG ftime /STATUS=outcome(1)
/STRATA=pair
/METHOD=ENTER exposici
/PRINT=CI(95).

* You can see that the odds-ratio and its 95%CI is the same with both
analyses .

Best regards,
Marta


       Inbox cluttering up with junk? Clean up with Yahoo! Mail.