Chi Square -- Crosstabs versus Nonparametric

[Ken Belzer]

Hi All,

I am wondering if someone can answer/explain the following to me.

I am running a 2 x 2 Pearson Chi square test with the following variables:
- Genotype?= + or yes (coded value =1) /- or no (coded value = 0)
- History of IV drug use/abuse?= + or yes(coded value =1) /- or no (coded value = 0)

I have used SPSS "Crosstabs" from the Descriptives menu in the past to derive Pearson chi square values and accompanying significance levels, but in version 15.0 there is an independent chi square?procedure under the "Nonparametric Tests" Menu.

I have run analyses using each method and my results?are conflicting.?Specifically, under the "Crosstabs" option,?my result is Pearson Chi-square (1) = 4.27, p = .023, wheras under the "Nonparametric" option?my result is?Pearson Chi-square (1) = 2.86, p = .091.

If it is of relevance my sample is as follows:
???????????????????????IV DRUG USE/ABUSE
??????????????????????????????????? NO / YES
????? GENOTYPE?--?NO???=? 19 /? = 7
???????????????????????? YES?=?? 15 /? = 0

Total N = 41.

Can anyone explain why the discrepant Pearson Chi Square?results and which is more valid and appropriate to report?

Thanks very much in advance for your feedback.
Ken

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

[cbautista]

Hi list,

I know that this question is not 100% related to SPSS, but maybe somebody
could provide any help on it!!!

I have a prevalence (or cross-sectional study) where two dependent
variables  (i.e. hepatitis B and hepatitis C) and three independent
variables have been measured (ie. sex, race, and drug use). In my
analysis, I have performed two logistic regression models:

hepatitis C = bo+b1(sex)+b2(race)+b3(drug use)
hepatitis B = bo+b4(sex)+b5(race)+b6(drug use)

in this study, the measurement (or independent variables) have been taken
simultaneously, but we are measuring two different outcomes.

So, I'd like to estimate a parameter for the difference in the effect of a
covariate (eg. sex) on hepatitis C and hepatiti B virus. Why Do I need
this? because, these differences could be informative.

any thoughs?

thanks in advance,

/Christian

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

[Anthony Babinec]

If I understand your table, the cells are

19 7
15 0

Is that correct?

You want to use the Crosstabs result, not the Npar Test
result. Also, because of the relatively small sample size,
you might look at the p-value associated with Fisher's
exact test.

Anthony Babinec
[hidden email]

"Fermentation and civilization are inseparable." John Ciardi
-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ken
Belzer
Sent: Friday, March 28, 2008 10:48 AM
To: [hidden email]
Subject: Chi Square -- Crosstabs versus Nonparametric

Hi All,

I am wondering if someone can answer/explain the following to me.

I am running a 2 x 2 Pearson Chi square test with the following variables:
- Genotype?= + or yes (coded value =1) /- or no (coded value = 0)
- History of IV drug use/abuse?= + or yes(coded value =1) /- or no (coded
value = 0)

I have used SPSS "Crosstabs" from the Descriptives menu in the past to
derive Pearson chi square values and accompanying significance levels, but
in version 15.0 there is an independent chi square?procedure under the
"Nonparametric Tests" Menu.

I have run analyses using each method and my results?are
conflicting.?Specifically, under the "Crosstabs" option,?my result is
Pearson Chi-square (1) = 4.27, p = .023, wheras under the "Nonparametric"
option?my result is?Pearson Chi-square (1) = 2.86, p = .091.

If it is of relevance my sample is as follows:
???????????????????????IV DRUG USE/ABUSE
??????????????????????????????????? NO / YES
????? GENOTYPE?--?NO???=? 19 /? = 7
???????????????????????? YES?=?? 15 /? = 0

Total N = 41.

Can anyone explain why the discrepant Pearson Chi Square?results and which
is more valid and appropriate to report?

Thanks very much in advance for your feedback.
Ken

=====================
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 Christian:

All data should be together in one dataset. You need a variable (called
hepatitis) with values 0=B and 1=C (or just the opposite, it doesn't mind).

Run a logistic model with all the predictors (sex, race and drug use)
plus the new variable hepatitis. To see if sex effect (for instance) is
equal in both hepatitis groups, add an interaction term between
hepatitis and sex. If it is significant, then the two betas are
statistically different.

HTH,
Marta

> Hi list,
>
> I know that this question is not 100% related to SPSS, but maybe somebody
> could provide any help on it!!!
>
> I have a prevalence (or cross-sectional study) where two dependent
> variables  (i.e. hepatitis B and hepatitis C) and three independent
> variables have been measured (ie. sex, race, and drug use). In my
> analysis, I have performed two logistic regression models:
>
> hepatitis C = bo+b1(sex)+b2(race)+b3(drug use)
> hepatitis B = bo+b4(sex)+b5(race)+b6(drug use)
>
> in this study, the measurement (or independent variables) have been taken
> simultaneously, but we are measuring two different outcomes.
>
> So, I'd like to estimate a parameter for the difference in the effect of a
> covariate (eg. sex) on hepatitis C and hepatiti B virus. Why Do I need
> this? because, these differences could be informative.
>
> any thoughs?
>
> thanks in advance,
>
> /Christian
>
> =====================
> 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 again

Reading my message (unfortunately, after I had hit "Send"), I realized
my explanation might be a bit misguiding, because I assumed that there
were two independent cross-sectional studies, but you have ONE study
with two outcomes, and there might even be cases with both diseases at
the same time. Now I see it is a rather complicated problem, because you
have somehow "paired" betas to test, not independent, as my previous
message assumed. Give me some time, and I will try to ask an
epidemiologist (who studied with Hosmer & Lemeshow time ago) at the
University about your problem. Although he is usually very busy, he owes
me some SPSS favors ;)

Best regards.
Marta

> Hi list,
>
> I know that this question is not 100% related to SPSS, but maybe somebody
> could provide any help on it!!!
>
> I have a prevalence (or cross-sectional study) where two dependent
> variables  (i.e. hepatitis B and hepatitis C) and three independent
> variables have been measured (ie. sex, race, and drug use). In my
> analysis, I have performed two logistic regression models:
>
> hepatitis C = bo+b1(sex)+b2(race)+b3(drug use)
> hepatitis B = bo+b4(sex)+b5(race)+b6(drug use)
>
> in this study, the measurement (or independent variables) have been taken
> simultaneously, but we are measuring two different outcomes.
>
> So, I'd like to estimate a parameter for the difference in the effect of a
> covariate (eg. sex) on hepatitis C and hepatiti B virus. Why Do I need
> this? because, these differences could be informative.
>
> any thoughs?
>
> thanks in advance,
>
> /Christian
>
> =====================
> 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 Christian:

My friend's reply.

z = [ln(OR1) -ln(OR2)] / [(SE12+SE22)0.5]

(Schlesselman JJ. Case-control studies. Design, conduct, analysis.
Oxford: Oxford University Press, 1982).

HTH,
Marta

> Hi list,
>
> I know that this question is not 100% related to SPSS, but maybe somebody
> could provide any help on it!!!
>
> I have a prevalence (or cross-sectional study) where two dependent
> variables  (i.e. hepatitis B and hepatitis C) and three independent
> variables have been measured (ie. sex, race, and drug use). In my
> analysis, I have performed two logistic regression models:
>
> hepatitis C = bo+b1(sex)+b2(race)+b3(drug use)
> hepatitis B = bo+b4(sex)+b5(race)+b6(drug use)
>
> in this study, the measurement (or independent variables) have been taken
> simultaneously, but we are measuring two different outcomes.
>
> So, I'd like to estimate a parameter for the difference in the effect of a
> covariate (eg. sex) on hepatitis C and hepatiti B virus. Why Do I need
> this? because, these differences could be informative.
>
> any thoughs?
>
> thanks in advance,
>
> /Christian
>
> =====================
> 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