Mantel-Haenszel Chi-Square or Alternative Test

[Glenys Lafrance]

Hello listers,

I am trying to discern group differences.  I am looking for a dichotomous
categorical independent-by-ordinal dependent test, when the ordinal
variable has a small number of categories (3-5) and/or some cells have
small counts.  In a previous post to this list (Mar. 15, 2006) a Mantel-
Haenszel chi-square was identified as taking into account ordinality of the
second variable.  Can SPSS compute a Mantel-Haenszel chi-square for a
rectangular table?  The option available with crosstabs appears to be
suited to 2x2 tables only (odds ratio).

Am I correct in eliminating Mann-Whitney U because apparently it is not
reliable unless the ordinal variable has a high number of categories and
similar distribution properties.

If there is a suitable test to consider, pls advise.  Thanks so much for
your consideration of this request.

Glenys Lafrance

[Marta García-Granero]

Hi Glenys

GL> I am trying to discern group differences.  I am looking for a dichotomous
GL> categorical independent-by-ordinal dependent test, when the ordinal
GL> variable has a small number of categories (3-5) and/or some cells have
GL> small counts.  In a previous post to this list (Mar. 15, 2006) a Mantel-
GL> Haenszel chi-square was identified as taking into account ordinality of the
GL> second variable.  Can SPSS compute a Mantel-Haenszel chi-square for a
GL> rectangular table?

Yes. It's part of the standard CROSSTABS output when you ask for
Chi-square statistics. It's called "Linear-by-linear association". If
you have small cell counts (observed or expected? it's not the same),
you had better ask for a exact p-value (provided you have the EXACT
TESTS module installed).

You can also use the Cochran-Armitage test for trend in this
situation (can be programmed with SPSS - I did it - but without the
exact p-value, only asymptotic significance)

GL>  The option available with crosstabs appears to be
GL> suited to 2x2 tables only (odds ratio).

You are mixing things a bit: this statistics you are talking about is
the weighed average of 2 or more stratum-OR when you are performing
stratified analysis. It has nothing to do with the situation we are
discussing now (trends).

GL> Am I correct in eliminating Mann-Whitney U because apparently it is not
GL> reliable unless the ordinal variable has a high number of categories and
GL> similar distribution properties.

Yes




--
Regards,
Dr. Marta García-Granero,PhD           mailto:[hidden email]
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)

[Margaret MacDougall]

Dear Marta

  I found your reply very interesting.

  Are you also able to advise why on occasions SPSS output for the Mantel-Haenszel chi-square test should include a footnote to the corresponding test statistic value which quotes a standardized value which is completely different from the tabulated one (please see my message of 17 October 2006).

  Many thanks

  Best wishes

  Margaret



Marta García-Granero <[hidden email]> wrote:
  Hi Glenys

GL> I am trying to discern group differences. I am looking for a dichotomous
GL> categorical independent-by-ordinal dependent test, when the ordinal
GL> variable has a small number of categories (3-5) and/or some cells have
GL> small counts. In a previous post to this list (Mar. 15, 2006) a Mantel-
GL> Haenszel chi-square was identified as taking into account ordinality of the
GL> second variable. Can SPSS compute a Mantel-Haenszel chi-square for a
GL> rectangular table?

Yes. It's part of the standard CROSSTABS output when you ask for
Chi-square statistics. It's called "Linear-by-linear association". If
you have small cell counts (observed or expected? it's not the same),
you had better ask for a exact p-value (provided you have the EXACT
TESTS module installed).

You can also use the Cochran-Armitage test for trend in this
situation (can be programmed with SPSS - I did it - but without the
exact p-value, only asymptotic significance)

GL> The option available with crosstabs appears to be
GL> suited to 2x2 tables only (odds ratio).

You are mixing things a bit: this statistics you are talking about is
the weighed average of 2 or more stratum-OR when you are performing
stratified analysis. It has nothing to do with the situation we are
discussing now (trends).

GL> Am I correct in eliminating Mann-Whitney U because apparently it is not
GL> reliable unless the ordinal variable has a high number of categories and
GL> similar distribution properties.

Yes




--
Regards,
Dr. Marta García-Granero,PhD mailto:[hidden email]
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)


 Send instant messages to your online friends http://uk.messenger.yahoo.com

[Anthony Babinec]

The linear-by-linear chi-square has 1 degree of freedom.
The standardized value looks like the square root of
the linear-by-linear chi-square. Recall that the
chi-square for 1 degree of freedom is a "z-square,"
where z is standard normal.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Margaret MacDougall
Sent: Wednesday, October 18, 2006 11:23 AM
To: [hidden email]
Subject: Re: Mantel-Haenszel Chi-Square or Alternative Test

Dear Marta

  I found your reply very interesting.

  Are you also able to advise why on occasions SPSS output for the
Mantel-Haenszel chi-square test should include a footnote to the
corresponding test statistic value which quotes a standardized value which
is completely different from the tabulated one (please see my message of 17
October 2006).

  Many thanks

  Best wishes

  Margaret



Marta García-Granero <[hidden email]> wrote:
  Hi Glenys

GL> I am trying to discern group differences. I am looking for a dichotomous
GL> categorical independent-by-ordinal dependent test, when the ordinal
GL> variable has a small number of categories (3-5) and/or some cells have
GL> small counts. In a previous post to this list (Mar. 15, 2006) a Mantel-
GL> Haenszel chi-square was identified as taking into account ordinality of
the
GL> second variable. Can SPSS compute a Mantel-Haenszel chi-square for a
GL> rectangular table?

Yes. It's part of the standard CROSSTABS output when you ask for
Chi-square statistics. It's called "Linear-by-linear association". If
you have small cell counts (observed or expected? it's not the same),
you had better ask for a exact p-value (provided you have the EXACT
TESTS module installed).

You can also use the Cochran-Armitage test for trend in this
situation (can be programmed with SPSS - I did it - but without the
exact p-value, only asymptotic significance)

GL> The option available with crosstabs appears to be
GL> suited to 2x2 tables only (odds ratio).

You are mixing things a bit: this statistics you are talking about is
the weighed average of 2 or more stratum-OR when you are performing
stratified analysis. It has nothing to do with the situation we are
discussing now (trends).

GL> Am I correct in eliminating Mann-Whitney U because apparently it is not
GL> reliable unless the ordinal variable has a high number of categories and
GL> similar distribution properties.

Yes




--
Regards,
Dr. Marta García-Granero,PhD mailto:[hidden email]
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)


 Send instant messages to your online friends http://uk.messenger.yahoo.com

[Margaret MacDougall]

Dear Anthony

  Thank you. What you are saying is very clear.

  Best wishes

  Margaret

Anthony Babinec <[hidden email]> wrote:
  The linear-by-linear chi-square has 1 degree of freedom.
The standardized value looks like the square root of
the linear-by-linear chi-square. Recall that the
chi-square for 1 degree of freedom is a "z-square,"
where z is standard normal.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Margaret MacDougall
Sent: Wednesday, October 18, 2006 11:23 AM
To: [hidden email]
Subject: Re: Mantel-Haenszel Chi-Square or Alternative Test

Dear Marta

I found your reply very interesting.

Are you also able to advise why on occasions SPSS output for the
Mantel-Haenszel chi-square test should include a footnote to the
corresponding test statistic value which quotes a standardized value which
is completely different from the tabulated one (please see my message of 17
October 2006).

Many thanks

Best wishes

Margaret



Marta García-Granero wrote:
Hi Glenys

GL> I am trying to discern group differences. I am looking for a dichotomous
GL> categorical independent-by-ordinal dependent test, when the ordinal
GL> variable has a small number of categories (3-5) and/or some cells have
GL> small counts. In a previous post to this list (Mar. 15, 2006) a Mantel-
GL> Haenszel chi-square was identified as taking into account ordinality of
the
GL> second variable. Can SPSS compute a Mantel-Haenszel chi-square for a
GL> rectangular table?

Yes. It's part of the standard CROSSTABS output when you ask for
Chi-square statistics. It's called "Linear-by-linear association". If
you have small cell counts (observed or expected? it's not the same),
you had better ask for a exact p-value (provided you have the EXACT
TESTS module installed).

You can also use the Cochran-Armitage test for trend in this
situation (can be programmed with SPSS - I did it - but without the
exact p-value, only asymptotic significance)

GL> The option available with crosstabs appears to be
GL> suited to 2x2 tables only (odds ratio).

You are mixing things a bit: this statistics you are talking about is
the weighed average of 2 or more stratum-OR when you are performing
stratified analysis. It has nothing to do with the situation we are
discussing now (trends).

GL> Am I correct in eliminating Mann-Whitney U because apparently it is not
GL> reliable unless the ordinal variable has a high number of categories and
GL> similar distribution properties.

Yes




--
Regards,
Dr. Marta García-Granero,PhD mailto:[hidden email]
Statistician

---
"It is unwise to use a statistical procedure whose use one does
not understand. SPSS syntax guide cannot supply this knowledge, and it
is certainly no substitute for the basic understanding of statistics
and statistical thinking that is essential for the wise choice of
methods and the correct interpretation of their results".

(Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind)


Send instant messages to your online friends http://uk.messenger.yahoo.com


 Send instant messages to your online friends http://uk.messenger.yahoo.com