compute logit and compare regression coef (linear and logistic)

[Iwona Pil]

i've a problem with logit transformation - i've dichotomous IDV (sex; 0 - M; 1 - F) and DV (0 - no; 1 - yes).
my results:

IDV = 0 and DV = 0 -> n = 20
IDV = 0 and DV = 1 -> n = 10
IDV = 1 and DV = 0 -> n = 5
IDV = 1 and DV = 1 -> n = 10

how to compute a logit on this results? if i'm right - if logit is correctly counted, non-standardized estimates in linear and logistic regression should be the same.

if anyone can help? :)

[Bruce Weaver]

Are you asking how to perform binary logistic regression using the summary data you show?  If so, read on.  If not, please clarify your question.


DATA LIST list / X Y Kount (3f5.0).
BEGIN DATA
0 0 20
0 1 10
1 0 5
1 1 10
END DATA.

WEIGHT by Kount.

LOGISTIC REGRESSION VARIABLES Y
  /METHOD=ENTER X
  /PRINT=CI(95)
.

* Compute same OR via CROSSTABS.
* To get the same result, temporarily
* recode 0 to 2 for both X and Y.

TEMPORARY.
RECODE X Y (0=2).
CROSSTABS X by Y / stat = risk.

Iwona Pil wrote

i've a problem with logit transformation - i've dichotomous IDV (sex; 0 - M; 1 - F) and DV (0 - no; 1 - yes).
my results:

IDV = 0 and DV = 0 -> n = 20
IDV = 0 and DV = 1 -> n = 10
IDV = 1 and DV = 0 -> n = 5
IDV = 1 and DV = 1 -> n = 10

how to compute a logit on this results? if i'm right - if logit is correctly counted, non-standardized estimates in linear and logistic regression should be the same.

if anyone can help? :)

[Iwona Pil]

not exact...
i'd like to transform my Y variable from dichotomous to linear.
i've dataset with N = 45 (sample database, attachment).
I tried to transform it, but my unstandarized coefficients  in linear and logistic regressions are not the same...

spinnfield_grayson.sav

[Bruce Weaver]

Pardon me if I'm being obtuse, but I don't understand.

1. What is a "linear" variable?  

2.  If the original variable has only two values, the transformed variable will also have two values, will it not?  (Two different values than the original variable, perhaps, but still only two values.)  

Iwona Pil wrote

not exact...
i'd like to transform my Y variable from dichotomous to linear.
i've dataset with N = 45 (sample database, attachment).
I tried to transform it, but my unstandarized coefficients  in linear and logistic regressions are not the same...

spinnfield_grayson.sav

[Rich Ulrich]

As I read it -- You are comparing results between using OLS with
computed Logits and using Maximum Likelihood Logistic Regression,
and are puzzled that they do not match exactly.

You are wrong in assuming that those two methods will give
identical results.  Logistic Regression, as commonly implemented,
uses maximum likelihood as the criterion, not least squares.  ML and
OLS give identical results for some simple problems, but not  for
this one.  A different criterion is being optimized.

One version of Logistic with Least Squares that I have read about was
in one of Finney's books.  It entailed appropriate weighting for variances,
and used an exact chi-squared for testing, rather than (default by
computer programs) F-tests.  If I recall correctly, this was labeled
Minimum Chi-squared logistic regression.  If you are using OLS on logits,
that will not match -- precisely -- either the ML or Finney's method that
improves on OLS.

Hope this is your concern. 

--
Rich Ulrich

> Date: Thu, 13 Feb 2014 22:38:24 -0800

> From: [hidden email]
> Subject: compute logit and compare regression coef (linear and logistic)
> To: [hidden email]
>
> i've a problem with logit transformation - i've dichotomous IDV (sex; 0 - M;
> 1 - F) and DV (0 - no; 1 - yes).
> my results:
>
> IDV = 0 and DV = 0 -> n = 20
> IDV = 0 and DV = 1 -> n = 10
> IDV = 1 and DV = 0 -> n = 5
> IDV = 1 and DV = 1 -> n = 10
>
> how to compute a logit on this results? if i'm right - if logit is correctly
> counted, non-standardized estimates in linear and logistic regression should
> be the same.
>
> if anyone can help? :)
>
>
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/compute-logit-and-compare-regression-coef-linear-and-logistic-tp5724493.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