## Logistic Regression (Hosmer-Lemeshow)

### Logistic Regression (Hosmer-Lemeshow)

Pages 63-64 in the SPSS Professional Statistics 7.5 book explain the H-L
test.

Jennifer R. Popovic
Statistician
Surveys and Research Staff

>----------

>Sent:  Tuesday, September 08, 1998 4:25 PM

>Subject:       Logistic Regression (Hosmer-Lemeshow)

>Can someone tell me where I can find out how SPSS (7.5) computes the
>Hosmer-Lemeshow goodness-of-fit statistics?  I can't quite reconcile what
>I am getting with H&L's description of the statistic.

>Thanks...

>+++++++++++++
>Rob Foss, Ph.D.
>Research Scientist
>UNC Highway Safety Research Center
>Chapel Hill, NC 27599-3430
>919-962-8702
>919-962-8710 (Fax)

### Logistic Regression (Hosmer-Lemeshow)

Can someone tell me where I can find out how SPSS (7.5) computes the
Hosmer-Lemeshow goodness-of-fit statistics?  I can't quite reconcile what
I am getting with H&L's description of the statistic.

Thanks...

+++++++++++++
Rob Foss, Ph.D.
Research Scientist
UNC Highway Safety Research Center
Chapel Hill, NC 27599-3430
919-962-8702
919-962-8710 (Fax)

Hello to all,

I'm curious about a difference in the Hosmer-Lemeshow
statistic when I run a hierarchical
(backward) stepwise logistic regression versus
running the final step of the individual blocks from the
just-described analysis separately.

Here's what I'm doing:

Method A: Hierarchical backward stepwise

logistic regression=nev_ever
/METHOD=BSTEP(LR)  premed_r agedevn pre pses famins hit g12_r casb
/CONTRAST (premed_r)=Indicator(1)
/method=bstep(lr) combat atroc threat malev inj peri
/method=bstep(lr)  emosshom insthome emossnow instnow strss pst recevent sevdep
/contrast (sevdep)=indicator(1)
/method=bstep(lr) treat
/contrast (treat)=indicator(1)
/print=goodfit ci(95) summary.

Method B:  Entering the variables retained at each of the 4 blocks
in the above analysis, done separately for each block.  (I was
doing this in order to get the predicted values for each block).

/* step 1 */
/* using the variables retained at block 1 in Method A analysis */

logistic regression var=nev_ever
/method=enter agedevn pre pses famins casb
/print goodfit
/save pred(st1_nev).

/* step 2 */

logistic regression var=nev_ever
/method=enter agedevn pre pses famins casb threat peri
/print=goodfit
/save pred(st2_nev).

/* step 3 */

logistic regression var=nev_ever
/method=enter agedevn pre pses famins casb threat peri emosshom pst sevdep
/contrast(sevdep)=indicator(1)
/print=goodfit
/save pred(st3_nev).

(no variables entered at step 4)

The sample size is identical in the final step, as is the rest of the
output (slopes, sig. tests, classification table)...I'm quite puzzled.