Conditional logistic problems

[Maguin, Eugene]

All,

I have been working with another list subscriber who has a matched (1:1)
dataset and is trying to analyze relationships between variables using
conditional logistic regression (CLR). I have no experience with this type
of model. However, I did find a posting by Marta (shown below) as well as
several discussion sites. Since spss logistic won't do a conditinal
analysis, the trick is to use the coxreg procedure to do so. I believe I
have faithfully followed Marta's directions but am getting no results and
so, there is something I don't understand. Per Marta's posting, I have

COXREG  ftime  /STATUS=outcome(1)  /STRATA=pair
  /METHOD=ENTER x /PRINT=CI(95).

Where
Ftime is the survival time variable computed such that ftime=outcome+1.
outcome is the case-control variable with 0=control, 1=case.
Pair is the strata variable, 161 pairs.
X is the IV.

My case processing box shows

Case Processing Summary
                                           N    Percent
Cases available in analysis     Event(a)        161     50.0%
                              Censored  0       .0%
                              Total     161     50.0%
Cases dropped     Cases with missing values     0       .0%
                  Cases with negative time      0       .0%
                  Censored cases before the
                earliest event in a stratum     161     50.0%
                          Total         322     100.0%
a       Dependent Variable: ftime

Variables in the Equation(b)
        Wald    df      Sig.
X       .       0(a)    .
aDegree of freedom reduced because of constant or linearly dependent
covariates
bConstant or Linearly Dependent Covariates
S = Stratum effect. x = .5093 + S;


If somebody can educate me about what I'm missing or need to look at, I'd
greatly appreciate it.

Thanks, Gene Maguin

=====================
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

[Reutter, Alex]

Gene,

This doesn't directly help your problems with COXREG, but if the dataset is matched 1:1 and not 1:many, you can alternatively use Multinomial Logistic Regression (NOMREG).  There's an example in Marija Norusis's Advanced Statistical Procedures Companion and another in the online help (Help>Case Studies, then Regression Models>Multinomial Logistic).

Cheers,
Alex


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: Monday, January 28, 2008 2:14 PM
To: [hidden email]
Subject: Conditional logistic problems

All,

I have been working with another list subscriber who has a matched (1:1)
dataset and is trying to analyze relationships between variables using
conditional logistic regression (CLR). I have no experience with this type
of model. However, I did find a posting by Marta (shown below) as well as
several discussion sites. Since spss logistic won't do a conditinal
analysis, the trick is to use the coxreg procedure to do so. I believe I
have faithfully followed Marta's directions but am getting no results and
so, there is something I don't understand. Per Marta's posting, I have

COXREG  ftime  /STATUS=outcome(1)  /STRATA=pair
  /METHOD=ENTER x /PRINT=CI(95).

Where
Ftime is the survival time variable computed such that ftime=outcome+1.
outcome is the case-control variable with 0=control, 1=case.
Pair is the strata variable, 161 pairs.
X is the IV.

My case processing box shows

Case Processing Summary
                                           N    Percent
Cases available in analysis     Event(a)        161     50.0%
                              Censored  0       .0%
                              Total     161     50.0%
Cases dropped     Cases with missing values     0       .0%
                  Cases with negative time      0       .0%
                  Censored cases before the
                earliest event in a stratum     161     50.0%
                          Total         322     100.0%
a       Dependent Variable: ftime

Variables in the Equation(b)
        Wald    df      Sig.
X       .       0(a)    .
aDegree of freedom reduced because of constant or linearly dependent
covariates
bConstant or Linearly Dependent Covariates
S = Stratum effect. x = .5093 + S;


If somebody can educate me about what I'm missing or need to look at, I'd
greatly appreciate it.

Thanks, Gene Maguin

=====================
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

[Marta Garcia-Granero]

Hi Gene:

See below a fully worked example, solved using the three approaches
(McNemar's OR with a MACRO, Mantel-Haenszel estimator and Conditional
logistic regression).

HTH,
Marta

(don't hesitate to ask me any question you might have concerning the syntax)

* Sample dataset (1:1 matched binary data, lower value means exposed in
both groups) *.
DATA LIST FREE/Case Control (2 F8).
BEGIN DATA
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 2 1 2
1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2
1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 2 1 2 1 2 1
2 1 2 1 2 1 2 1 2 1 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2
END DATA.
VALUE LABEL Case Control 1'Exposed to Risk' 2 'Non exposed'.
LIST /CASES=FROM 1 TO 10.

* 1rst approach, valid only for 1:1 matched data, using a MACRO *.
* MACRO call (see my message to the list: "Re: Help with Matrix output
(& Paired data OR macro too, for H)"*.

PAIREDOR Case Control.

* 2nd approach: Mantel-Haenszel stratified analysis (valid for 1:1 or
1:k matched data) *.

* First of all, data need restructuring (using VARSTOCASES) *.
VARSTOCASES
 /ID = pair
 /MAKE Exposition 'Exposed to risk factor' FROM Case Control
 /INDEX = Outcome(2).
VALUE LABEL Outcome 1'Case' 2 'Control'.

* MH OR (ignore all tables but the last) *.
CROSSTABS
  /TABLES=Exposition  BY Outcome  BY pair
  /FORMAT=NOTABLES
  /STATISTIC=CMH(1)
  /COUNT ROUND CELL .

* 3rd approach: Conditional logistic regression (thru Cox regression) *.
* We will start from the dataset we have just restructured *.
* We need three variables:
- Exposition -> it is already OK
- Pseudotime: should have these two values: 1 for cases and 2 for controls
  we already have it (named outcome).
- Status: should have these two values: 0 for controls and 1 for cases
  we must create it.

COMPUTE status=(outcome=1).

COXREG
  Outcome  /STATUS=status(1)  /STRATA=pair
  /CONTRAST (Exposition)=Indicator
  /METHOD=ENTER Exposition
  /PRINT=CI(95).

The three methods give the same result (as expected).

> All,
>
> I have been working with another list subscriber who has a matched (1:1)
> dataset and is trying to analyze relationships between variables using
> conditional logistic regression (CLR). I have no experience with this type
> of model. However, I did find a posting by Marta (shown below) as well as
> several discussion sites. Since spss logistic won't do a conditinal
> analysis, the trick is to use the coxreg procedure to do so. I believe I
> have faithfully followed Marta's directions but am getting no results and
> so, there is something I don't understand. Per Marta's posting, I have
>
> COXREG  ftime  /STATUS=outcome(1)  /STRATA=pair
>   /METHOD=ENTER x /PRINT=CI(95).
>
> Where
> Ftime is the survival time variable computed such that ftime=outcome+1.
> outcome is the case-control variable with 0=control, 1=case.
> Pair is the strata variable, 161 pairs.
> X is the IV.
>
> My case processing box shows
>
> Case Processing Summary
>                                            N    Percent
> Cases available in analysis     Event(a)        161     50.0%
>                               Censored  0       .0%
>                               Total     161     50.0%
> Cases dropped     Cases with missing values     0       .0%
>                   Cases with negative time      0       .0%
>                   Censored cases before the
>                 earliest event in a stratum     161     50.0%
>                           Total         322     100.0%
> a       Dependent Variable: ftime
>
> Variables in the Equation(b)
>         Wald    df      Sig.
> X       .       0(a)    .
> aDegree of freedom reduced because of constant or linearly dependent
> covariates
> bConstant or Linearly Dependent Covariates
> S = Stratum effect. x = .5093 + S;
>
>
> If somebody can educate me about what I'm missing or need to look at, I'd
> greatly appreciate it.
>
> Thanks, Gene Maguin
>
> =====================
> 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

[Maguin, Eugene]

Alex,

Thank you for pointing that out to me. After looking at the presentation, it
occurs to me that dichotomous logistic regression could be used as well.
Would that be true?

Gene Maguin

=====================
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