SPSSX Discussion

Interaction Between Categorical Variables_Another Try?

Classic

List

Threaded

10 messages Options

Chao Yawo

Interaction Between Categorical Variables_Another Try?

Hello,

Thanks for all your responses regarding my previous question. I
certainly plan to take a quick tutorial on HLM, and re-analyze my
data using such a model. I believe it would be more instructive.

In the meantime, I have re-run my logistic regression model
(predicting the odds of desiring an HIV test) with the following
covariates.

-- x1(categorical) =1 if a respondent has used a condom or not during
last sexual intercourse, and 0 if not

-- I collapsed the proportion of a respondent's community holding
stigmatizing attitudes into tertiles corresponding to low, medium and
high community stigma, resulting into two dummy variables, X2 & X3
(with the medium category being referenced).

-- To test my hypothesis that the effect of risky unprotected sex on
desire for an HIV test could differ by levels of community stigma, an
interaction term
x1 by x2 (RiskySex_LowCommunityStigma)
x1 by x3 (RiskySex_HighCommunityStigma) were included:

One of the terms is curiously and surprisingly positive. Relevant
sections of the output are as follows:

Variable --> Odds ratio --> 95%CI ---> B-coef
x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
constant --> 1.36 --> 0.63 - 2.97 ---> 0.31

I am curious about the RiskySex_LowCommunityStigma interaction -- its
odds ratio is 3.78. The positive interaction indicates that the
positive effect of
engaging on risky sex is greater among individuals in more tolerant
communities (i.e., those in communities with lower proportion of
individuals holding stigmatizing attitudes). Put differently,
individuals who live in less tolerant (i.e., those with higher
proportion of individuals holding stigmatizing attitudes) results in
more of a disadvantage for respondents who have engaged in risky sex.

Is this interpretation about correct.... ?

Thanks again for all your help, Cy

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

Bruce Weaver

Re: Interaction Between Categorical Variables_Another Try?

Administrator

Two comments:

1. Your table of results is missing the odds ratio for X3 and the X1*X3 product term.

2. To understand the nature of the interaction, work out fitted values of Y at the 6 combinations of condom used and attitude, and plot them.

For point 2, I like to use the log-odds scale, because on that scale, things that are linear *look* linear. If you plot fitted odds or probabilities, things that are linear on the log-odds scale no longer look linear, which makes it harder to see the nature of the interaction.

Once you have the plot described above, you will be able to see more easily that the coefficients for the product terms give you differences of differences on the log-odds scale, or ratios of odds ratios on the odds scale. (See James Jaccard's Sage monograph on interactions in logistic regression.)

Chao Yawo wrote

Hello,

Thanks for all your responses regarding my previous question. I
certainly plan to take a quick tutorial on HLM, and re-analyze my
data using such a model. I believe it would be more instructive.

In the meantime, I have re-run my logistic regression model
(predicting the odds of desiring an HIV test) with the following
covariates.

-- x1(categorical) =1 if a respondent has used a condom or not during
last sexual intercourse, and 0 if not

-- I collapsed the proportion of a respondent's community holding
stigmatizing attitudes into tertiles corresponding to low, medium and
high community stigma, resulting into two dummy variables, X2 & X3
(with the medium category being referenced).

-- To test my hypothesis that the effect of risky unprotected sex on
desire for an HIV test could differ by levels of community stigma, an
interaction term
x1 by x2 (RiskySex_LowCommunityStigma)
x1 by x3 (RiskySex_HighCommunityStigma) were included:

One of the terms is curiously and surprisingly positive. Relevant
sections of the output are as follows:

Variable --> Odds ratio --> 95%CI ---> B-coef
x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
constant --> 1.36 --> 0.63 - 2.97 ---> 0.31

I am curious about the RiskySex_LowCommunityStigma interaction -- its
odds ratio is 3.78. The positive interaction indicates that the
positive effect of
engaging on risky sex is greater among individuals in more tolerant
communities (i.e., those in communities with lower proportion of
individuals holding stigmatizing attitudes). Put differently,
individuals who live in less tolerant (i.e., those with higher
proportion of individuals holding stigmatizing attitudes) results in
more of a disadvantage for respondents who have engaged in risky sex.

Is this interpretation about correct.... ?

Thanks again for all your help, Cy

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (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

--
Bruce Weaver
bweaver@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

PLEASE NOTE THE FOLLOWING:
1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above.
2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/).

Ryan

Re: Interaction Between Categorical Variables_Another Try?

In reply to this post by Chao Yawo

Chao,

Generally it is not a good idea to convert a continous variable into a categorical variable before analysis.

Ryan

Chao Yawo wrote

Hello,

Thanks for all your responses regarding my previous question. I
certainly plan to take a quick tutorial on HLM, and re-analyze my
data using such a model. I believe it would be more instructive.

In the meantime, I have re-run my logistic regression model
(predicting the odds of desiring an HIV test) with the following
covariates.

-- x1(categorical) =1 if a respondent has used a condom or not during
last sexual intercourse, and 0 if not

-- I collapsed the proportion of a respondent's community holding
stigmatizing attitudes into tertiles corresponding to low, medium and
high community stigma, resulting into two dummy variables, X2 & X3
(with the medium category being referenced).

-- To test my hypothesis that the effect of risky unprotected sex on
desire for an HIV test could differ by levels of community stigma, an
interaction term
x1 by x2 (RiskySex_LowCommunityStigma)
x1 by x3 (RiskySex_HighCommunityStigma) were included:

One of the terms is curiously and surprisingly positive. Relevant
sections of the output are as follows:

Variable --> Odds ratio --> 95%CI ---> B-coef
x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
constant --> 1.36 --> 0.63 - 2.97 ---> 0.31

I am curious about the RiskySex_LowCommunityStigma interaction -- its
odds ratio is 3.78. The positive interaction indicates that the
positive effect of
engaging on risky sex is greater among individuals in more tolerant
communities (i.e., those in communities with lower proportion of
individuals holding stigmatizing attitudes). Put differently,
individuals who live in less tolerant (i.e., those with higher
proportion of individuals holding stigmatizing attitudes) results in
more of a disadvantage for respondents who have engaged in risky sex.

Is this interpretation about correct.... ?

Thanks again for all your help, Cy

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (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

Ryan

Re: Interaction Between Categorical Variables_Another Try?

Hello:

I was asked to post my thinking behind the statement I made about my general reluctance to convert a continuous variable into a categorical variable before analysis. The fundamental reason I avoid this tactic is because of loss of information. There are certainly situations where it would be appropriate, but as a general rule of thumb, I try to avoid removing information, as that information might turn out to be critical.

Best,

Ryan

rblack wrote

Chao,

Generally it is not a good idea to convert a continous variable into a categorical variable before analysis.

Ryan

Chao Yawo wrote

Hello,

Thanks for all your responses regarding my previous question. I
certainly plan to take a quick tutorial on HLM, and re-analyze my
data using such a model. I believe it would be more instructive.

In the meantime, I have re-run my logistic regression model
(predicting the odds of desiring an HIV test) with the following
covariates.

-- x1(categorical) =1 if a respondent has used a condom or not during
last sexual intercourse, and 0 if not

-- I collapsed the proportion of a respondent's community holding
stigmatizing attitudes into tertiles corresponding to low, medium and
high community stigma, resulting into two dummy variables, X2 & X3
(with the medium category being referenced).

-- To test my hypothesis that the effect of risky unprotected sex on
desire for an HIV test could differ by levels of community stigma, an
interaction term
x1 by x2 (RiskySex_LowCommunityStigma)
x1 by x3 (RiskySex_HighCommunityStigma) were included:

One of the terms is curiously and surprisingly positive. Relevant
sections of the output are as follows:

Variable --> Odds ratio --> 95%CI ---> B-coef
x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
constant --> 1.36 --> 0.63 - 2.97 ---> 0.31

I am curious about the RiskySex_LowCommunityStigma interaction -- its
odds ratio is 3.78. The positive interaction indicates that the
positive effect of
engaging on risky sex is greater among individuals in more tolerant
communities (i.e., those in communities with lower proportion of
individuals holding stigmatizing attitudes). Put differently,
individuals who live in less tolerant (i.e., those with higher
proportion of individuals holding stigmatizing attitudes) results in
more of a disadvantage for respondents who have engaged in risky sex.

Is this interpretation about correct.... ?

Thanks again for all your help, Cy

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (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

Bruce Weaver

Re: Interaction Between Categorical Variables_Another Try?

Administrator

There are several publications that make this point. One very readable one is "Breaking Up is Hard to Do", by David Streiner (mind the line-wrap):

http://server03.cpa-apc.org:8080/Publications/archives/CJP/2002/april/researchMethodsDichotomizingData.asp

rblack wrote

Hello:

I was asked to post my thinking behind the statement I made about my general reluctance to convert a continuous variable into a categorical variable before analysis. The fundamental reason I avoid this tactic is because of loss of information. There are certainly situations where it would be appropriate, but as a general rule of thumb, I try to avoid removing information, as that information might turn out to be critical.

Best,

Ryan

rblack wrote

Chao,

Generally it is not a good idea to convert a continous variable into a categorical variable before analysis.

Ryan

Chao Yawo wrote

Hello,

Thanks for all your responses regarding my previous question. I
certainly plan to take a quick tutorial on HLM, and re-analyze my
data using such a model. I believe it would be more instructive.

In the meantime, I have re-run my logistic regression model
(predicting the odds of desiring an HIV test) with the following
covariates.

-- x1(categorical) =1 if a respondent has used a condom or not during
last sexual intercourse, and 0 if not

-- I collapsed the proportion of a respondent's community holding
stigmatizing attitudes into tertiles corresponding to low, medium and
high community stigma, resulting into two dummy variables, X2 & X3
(with the medium category being referenced).

-- To test my hypothesis that the effect of risky unprotected sex on
desire for an HIV test could differ by levels of community stigma, an
interaction term
x1 by x2 (RiskySex_LowCommunityStigma)
x1 by x3 (RiskySex_HighCommunityStigma) were included:

One of the terms is curiously and surprisingly positive. Relevant
sections of the output are as follows:

Variable --> Odds ratio --> 95%CI ---> B-coef
x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
constant --> 1.36 --> 0.63 - 2.97 ---> 0.31

I am curious about the RiskySex_LowCommunityStigma interaction -- its
odds ratio is 3.78. The positive interaction indicates that the
positive effect of
engaging on risky sex is greater among individuals in more tolerant
communities (i.e., those in communities with lower proportion of
individuals holding stigmatizing attitudes). Put differently,
individuals who live in less tolerant (i.e., those with higher
proportion of individuals holding stigmatizing attitudes) results in
more of a disadvantage for respondents who have engaged in risky sex.

Is this interpretation about correct.... ?

Thanks again for all your help, Cy

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (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

statisticsdoc

Re: Interaction Between Categorical Variables_Another Try?

Chao,
Why did you divide the continuous measure of stigma into three categorical levels? Are you expecting to see a non-linear.interaction (say an effect that occurs when condom use is combined with high stigma but not with low or medium stigma). If this is the case, you would be in better shape entering a quadratic as well as a linear term for the continuous measure of stigna.
Best (ie square the scores on stigma).
HTH
Steve Brand

www.StatisticsDoc.com

www.StatisticsDoc.com

-----Original Message-----
From: Bruce Weaver <[hidden email]>
Date: Wed, 31 Mar 2010 12:26:03
To: <[hidden email]>
Subject: Re: Interaction Between Categorical Variables_Another Try?

There are several publications that make this point. One very readable one
is "Breaking Up is Hard to Do", by David Streiner (mind the line-wrap):

http://server03.cpa-apc.org:8080/Publications/archives/CJP/2002/april/researchMethodsDichotomizingData.asp

rblack wrote:

>
> Hello:
>
> I was asked to post my thinking behind the statement I made about my
> general reluctance to convert a continuous variable into a categorical
> variable before analysis. The fundamental reason I avoid this tactic is
> because of loss of information. There are certainly situations where it
> would be appropriate, but as a general rule of thumb, I try to avoid
> removing information, as that information might turn out to be critical.
>
> Best,
>
> Ryan
>
>
> rblack wrote:
>>
>> Chao,
>>
>> Generally it is not a good idea to convert a continous variable into a
>> categorical variable before analysis.
>>
>> Ryan
>>
>>
>> Chao Yawo wrote:
>>>
>>> Hello,
>>>
>>> Thanks for all your responses regarding my previous question. I
>>> certainly plan to take a quick tutorial on HLM, and re-analyze my
>>> data using such a model. I believe it would be more instructive.
>>>
>>> In the meantime, I have re-run my logistic regression model
>>> (predicting the odds of desiring an HIV test) with the following
>>> covariates.
>>>
>>> -- x1(categorical) =1 if a respondent has used a condom or not during
>>> last sexual intercourse, and 0 if not
>>>
>>> -- I collapsed the proportion of a respondent's community holding
>>> stigmatizing attitudes into tertiles corresponding to low, medium and
>>> high community stigma, resulting into two dummy variables, X2 & X3
>>> (with the medium category being referenced).
>>>
>>> -- To test my hypothesis that the effect of risky unprotected sex on
>>> desire for an HIV test could differ by levels of community stigma, an
>>> interaction term
>>> x1 by x2 (RiskySex_LowCommunityStigma)
>>> x1 by x3 (RiskySex_HighCommunityStigma) were included:
>>>
>>> One of the terms is curiously and surprisingly positive. Relevant
>>> sections of the output are as follows:
>>>
>>> Variable --> Odds ratio --> 95%CI ---> B-coef
>>> x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
>>> x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
>>> x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
>>> constant --> 1.36 --> 0.63 - 2.97 ---> 0.31
>>>
>>> I am curious about the RiskySex_LowCommunityStigma interaction -- its
>>> odds ratio is 3.78. The positive interaction indicates that the
>>> positive effect of
>>> engaging on risky sex is greater among individuals in more tolerant
>>> communities (i.e., those in communities with lower proportion of
>>> individuals holding stigmatizing attitudes). Put differently,
>>> individuals who live in less tolerant (i.e., those with higher
>>> proportion of individuals holding stigmatizing attitudes) results in
>>> more of a disadvantage for respondents who have engaged in risky sex.
>>>
>>> Is this interpretation about correct.... ?
>>>
>>> Thanks again for all your help, Cy
>>>
>>> =====================
>>> 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
>>>
>>>
>>
>>
>
>

-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/
"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.
--
View this message in context: http://old.nabble.com/Interaction-Between-Categorical-Variables_Another-Try--tp28076039p28100684.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

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

James Whanger

Re: Interaction Between Categorical Variables_Another Try?

Steve,

What is your rationale for assuming that an ordered categorical variable posited to measure "stigma" is in fact a continuous variable?

Regards,

Jim

On Wed, Mar 31, 2010 at 4:12 PM, Statisticsdoc Consulting <[hidden email]> wrote:

Chao,
Why did you divide the continuous measure of stigma into three categorical levels? Are you expecting to see a non-linear.interaction (say an effect that occurs when condom use is combined with high stigma but not with low or medium stigma). If this is the case, you would be in better shape entering a quadratic as well as a linear term for the continuous measure of stigna.
Best (ie square the scores on stigma).
HTH
Steve Brand

www.StatisticsDoc.com

www.StatisticsDoc.com

-----Original Message-----
From: Bruce Weaver <[hidden email]>
Date: Wed, 31 Mar 2010 12:26:03
To: <[hidden email]>
Subject: Re: Interaction Between Categorical Variables_Another Try?

There are several publications that make this point. One very readable one
is "Breaking Up is Hard to Do", by David Streiner (mind the line-wrap):

http://server03.cpa-apc.org:8080/Publications/archives/CJP/2002/april/researchMethodsDichotomizingData.asp

rblack wrote:
>
> Hello:
>
> I was asked to post my thinking behind the statement I made about my
> general reluctance to convert a continuous variable into a categorical
> variable before analysis. The fundamental reason I avoid this tactic is
> because of loss of information. There are certainly situations where it
> would be appropriate, but as a general rule of thumb, I try to avoid
> removing information, as that information might turn out to be critical.
>

> Best,
>
> Ryan
>
>
> rblack wrote:
>>
>> Chao,
>>
>> Generally it is not a good idea to convert a continous variable into a
>> categorical variable before analysis.
>>
>> Ryan
>>
>>
>> Chao Yawo wrote:
>>>
>>> Hello,
>>>
>>> Thanks for all your responses regarding my previous question. I
>>> certainly plan to take a quick tutorial on HLM, and re-analyze my
>>> data using such a model. I believe it would be more instructive.
>>>
>>> In the meantime, I have re-run my logistic regression model
>>> (predicting the odds of desiring an HIV test) with the following
>>> covariates.
>>>
>>> -- x1(categorical) =1 if a respondent has used a condom or not during
>>> last sexual intercourse, and 0 if not
>>>
>>> -- I collapsed the proportion of a respondent's community holding
>>> stigmatizing attitudes into tertiles corresponding to low, medium and
>>> high community stigma, resulting into two dummy variables, X2 & X3
>>> (with the medium category being referenced).
>>>
>>> -- To test my hypothesis that the effect of risky unprotected sex on
>>> desire for an HIV test could differ by levels of community stigma, an
>>> interaction term
>>> x1 by x2 (RiskySex_LowCommunityStigma)
>>> x1 by x3 (RiskySex_HighCommunityStigma) were included:
>>>
>>> One of the terms is curiously and surprisingly positive. Relevant
>>> sections of the output are as follows:
>>>
>>> Variable --> Odds ratio --> 95%CI ---> B-coef
>>> x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
>>> x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
>>> x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
>>> constant --> 1.36 --> 0.63 - 2.97 ---> 0.31
>>>
>>> I am curious about the RiskySex_LowCommunityStigma interaction -- its
>>> odds ratio is 3.78. The positive interaction indicates that the
>>> positive effect of
>>> engaging on risky sex is greater among individuals in more tolerant
>>> communities (i.e., those in communities with lower proportion of
>>> individuals holding stigmatizing attitudes). Put differently,
>>> individuals who live in less tolerant (i.e., those with higher
>>> proportion of individuals holding stigmatizing attitudes) results in
>>> more of a disadvantage for respondents who have engaged in risky sex.
>>>
>>> Is this interpretation about correct.... ?
>>>
>>> Thanks again for all your help, Cy
>>>
>>> =====================
>>> 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
>>>
>>>
>>
>>
>
>

-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/
"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.
--
View this message in context: http://old.nabble.com/Interaction-Between-Categorical-Variables_Another-Try--tp28076039p28100684.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

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

statisticsdoc

Re: Interaction Between Categorical Variables_Another Try?

As a number of posters have noted, Chao originally measured stigma as a continuous variable and then recoded ranges of scores into three categories.
HTH
Steve Brand

www.StatisticsDoc.com

From: "James C. Whanger" <[hidden email]>

Date: Wed, 31 Mar 2010 17:12:48 -0400

To: <[hidden email]>

Subject: Re: Interaction Between Categorical Variables_Another Try?

Steve,

What is your rationale for assuming that an ordered categorical variable posited to measure "stigma" is in fact a continuous variable?

Regards,

Jim

On Wed, Mar 31, 2010 at 4:12 PM, Statisticsdoc Consulting <[hidden email]> wrote:

Chao,
Why did you divide the continuous measure of stigma into three categorical levels? Are you expecting to see a non-linear.interaction (say an effect that occurs when condom use is combined with high stigma but not with low or medium stigma). If this is the case, you would be in better shape entering a quadratic as well as a linear term for the continuous measure of stigna.
Best (ie square the scores on stigma).
HTH
Steve Brand

www.StatisticsDoc.com

www.StatisticsDoc.com

-----Original Message-----
From: Bruce Weaver <[hidden email]>
Date: Wed, 31 Mar 2010 12:26:03
To: <[hidden email]>
Subject: Re: Interaction Between Categorical Variables_Another Try?

There are several publications that make this point. One very readable one
is "Breaking Up is Hard to Do", by David Streiner (mind the line-wrap):

http://server03.cpa-apc.org:8080/Publications/archives/CJP/2002/april/researchMethodsDichotomizingData.asp

rblack wrote:
>
> Hello:
>
> I was asked to post my thinking behind the statement I made about my
> general reluctance to convert a continuous variable into a categorical
> variable before analysis. The fundamental reason I avoid this tactic is
> because of loss of information. There are certainly situations where it
> would be appropriate, but as a general rule of thumb, I try to avoid
> removing information, as that information might turn out to be critical.
>

> Best,
>
> Ryan
>
>
> rblack wrote:
>>
>> Chao,
>>
>> Generally it is not a good idea to convert a continous variable into a
>> categorical variable before analysis.
>>
>> Ryan
>>
>>
>> Chao Yawo wrote:
>>>
>>> Hello,
>>>
>>> Thanks for all your responses regarding my previous question. I
>>> certainly plan to take a quick tutorial on HLM, and re-analyze my
>>> data using such a model. I believe it would be more instructive.
>>>
>>> In the meantime, I have re-run my logistic regression model
>>> (predicting the odds of desiring an HIV test) with the following
>>> covariates.
>>>
>>> -- x1(categorical) =1 if a respondent has used a condom or not during
>>> last sexual intercourse, and 0 if not
>>>
>>> -- I collapsed the proportion of a respondent's community holding
>>> stigmatizing attitudes into tertiles corresponding to low, medium and
>>> high community stigma, resulting into two dummy variables, X2 & X3
>>> (with the medium category being referenced).
>>>
>>> -- To test my hypothesis that the effect of risky unprotected sex on
>>> desire for an HIV test could differ by levels of community stigma, an
>>> interaction term
>>> x1 by x2 (RiskySex_LowCommunityStigma)
>>> x1 by x3 (RiskySex_HighCommunityStigma) were included:
>>>
>>> One of the terms is curiously and surprisingly positive. Relevant
>>> sections of the output are as follows:
>>>
>>> Variable --> Odds ratio --> 95%CI ---> B-coef
>>> x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
>>> x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
>>> x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
>>> constant --> 1.36 --> 0.63 - 2.97 ---> 0.31
>>>
>>> I am curious about the RiskySex_LowCommunityStigma interaction -- its
>>> odds ratio is 3.78. The positive interaction indicates that the
>>> positive effect of
>>> engaging on risky sex is greater among individuals in more tolerant
>>> communities (i.e., those in communities with lower proportion of
>>> individuals holding stigmatizing attitudes). Put differently,
>>> individuals who live in less tolerant (i.e., those with higher
>>> proportion of individuals holding stigmatizing attitudes) results in
>>> more of a disadvantage for respondents who have engaged in risky sex.
>>>
>>> Is this interpretation about correct.... ?
>>>
>>> Thanks again for all your help, Cy
>>>
>>> =====================
>>> 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
>>>
>>>
>>
>>
>
>

-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/
"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.
--
View this message in context: http://old.nabble.com/Interaction-Between-Categorical-Variables_Another-Try--tp28076039p28100684.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

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

Chao Yawo

Re: Interaction Between Categorical Variables_Another Try?

Thanks --- Bruce, Ryan,and Steve,

Yes, I originally had a continuous variable that measures the
proportion of a respondent's community that hold certain stigmatizing
view (this was calculated by substracting the respondent's core from
the total score of all other respondents in the same stratum as the
respondent). I know an HLM model can handle this question better -
given the nested structure of the data. I plan to do it once I finish
going through some tutorials.

My expecting is extent of stigma (especially community level stigma)
will moderate the effect of risky behavior such as condom use on
desire to get HIV test. So, if stigma is high, then it would be less
likely for people to seek to test for HIV, even if they are engaged in
risky behaviors.

However, the continuous variable is a bit skewed, so I thought of
categorizing into three variables. And frankly, I was under the
impression that the categorical by categorical interaction was easier
to interpret than the categorical by continuous interaction effect
(but I may be wrong).

Thanks - Chao

==========
On Wed, Mar 31, 2010 at 7:43 PM, Statisticsdoc Consulting
<[hidden email]> wrote:

> As a number of posters have noted, Chao originally measured stigma as a
> continuous variable and then recoded ranges of scores into three categories.
> HTH
> Steve Brand
>
> www.StatisticsDoc.com
>
> ________________________________
> From: "James C. Whanger" <[hidden email]>
> Date: Wed, 31 Mar 2010 17:12:48 -0400
> To: <[hidden email]>
> Subject: Re: Interaction Between Categorical Variables_Another Try?
> Steve,
>
> What is your rationale for assuming that an ordered categorical variable
> posited to measure "stigma" is in fact a continuous variable?
>
> Regards,
>
> Jim
>
> On Wed, Mar 31, 2010 at 4:12 PM, Statisticsdoc Consulting
> <[hidden email]> wrote:
>>
>> Chao,
>> Why did you divide the continuous measure of stigma into three categorical
>> levels? Are you expecting to see a non-linear.interaction (say an effect
>> that occurs when condom use is combined with high stigma but not with low or
>> medium stigma). If this is the case, you would be in better shape entering
>> a quadratic as well as a linear term for the continuous measure of stigna.
>> Best (ie square the scores on stigma).
>> HTH
>> Steve Brand
>>
>> www.StatisticsDoc.com
>>
>>
>> www.StatisticsDoc.com
>>
>> -----Original Message-----
>> From: Bruce Weaver <[hidden email]>
>> Date: Wed, 31 Mar 2010 12:26:03
>> To: <[hidden email]>
>> Subject: Re: Interaction Between Categorical Variables_Another Try?
>>
>> There are several publications that make this point. One very readable
>> one
>> is "Breaking Up is Hard to Do", by David Streiner (mind the line-wrap):
>>
>>
>>
>> http://server03.cpa-apc.org:8080/Publications/archives/CJP/2002/april/researchMethodsDichotomizingData.asp
>>
>>
>>
>>
>> rblack wrote:
>> >
>> > Hello:
>> >
>> > I was asked to post my thinking behind the statement I made about my
>> > general reluctance to convert a continuous variable into a categorical
>> > variable before analysis. The fundamental reason I avoid this tactic is
>> > because of loss of information. There are certainly situations where it
>> > would be appropriate, but as a general rule of thumb, I try to avoid
>> > removing information, as that information might turn out to be critical.
>> >
>> > Best,
>> >
>> > Ryan
>> >
>> >
>> > rblack wrote:
>> >>
>> >> Chao,
>> >>
>> >> Generally it is not a good idea to convert a continous variable into a
>> >> categorical variable before analysis.
>> >>
>> >> Ryan
>> >>
>> >>
>> >> Chao Yawo wrote:
>> >>>
>> >>> Hello,
>> >>>
>> >>> Thanks for all your responses regarding my previous question. I
>> >>> certainly plan to take a quick tutorial on HLM, and re-analyze my
>> >>> data using such a model. I believe it would be more instructive.
>> >>>
>> >>> In the meantime, I have re-run my logistic regression model
>> >>> (predicting the odds of desiring an HIV test) with the following
>> >>> covariates.
>> >>>
>> >>> -- x1(categorical) =1 if a respondent has used a condom or not during
>> >>> last sexual intercourse, and 0 if not
>> >>>
>> >>> -- I collapsed the proportion of a respondent's community holding
>> >>> stigmatizing attitudes into tertiles corresponding to low, medium and
>> >>> high community stigma, resulting into two dummy variables, X2 & X3
>> >>> (with the medium category being referenced).
>> >>>
>> >>> -- To test my hypothesis that the effect of risky unprotected sex on
>> >>> desire for an HIV test could differ by levels of community stigma, an
>> >>> interaction term
>> >>> x1 by x2 (RiskySex_LowCommunityStigma)
>> >>> x1 by x3 (RiskySex_HighCommunityStigma) were included:
>> >>>
>> >>> One of the terms is curiously and surprisingly positive. Relevant
>> >>> sections of the output are as follows:
>> >>>
>> >>> Variable --> Odds ratio --> 95%CI ---> B-coef
>> >>> x1 --> 2.53 --> 1.01 - 6.34 ---> 0.93
>> >>> x2 --> 1.31 --> 0.71 - 2.39 ---> 0.27
>> >>> x1 by x2 --> 3.78 --> 1.47 - 9.72 ---> 1.33
>> >>> constant --> 1.36 --> 0.63 - 2.97 ---> 0.31
>> >>>
>> >>> I am curious about the RiskySex_LowCommunityStigma interaction -- its
>> >>> odds ratio is 3.78. The positive interaction indicates that the
>> >>> positive effect of
>> >>> engaging on risky sex is greater among individuals in more tolerant
>> >>> communities (i.e., those in communities with lower proportion of
>> >>> individuals holding stigmatizing attitudes). Put differently,
>> >>> individuals who live in less tolerant (i.e., those with higher
>> >>> proportion of individuals holding stigmatizing attitudes) results in
>> >>> more of a disadvantage for respondents who have engaged in risky sex.
>> >>>
>> >>> Is this interpretation about correct.... ?
>> >>>
>> >>> Thanks again for all your help, Cy
>> >>>
>> >>> =====================
>> >>> 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
>> >>>
>> >>>
>> >>
>> >>
>> >
>> >
>>
>>
>> -----
>> --
>> Bruce Weaver
>> [hidden email]
>> http://sites.google.com/a/lakeheadu.ca/bweaver/
>> "When all else fails, RTFM."
>>
>> NOTE: My Hotmail account is not monitored regularly.
>> To send me an e-mail, please use the address shown above.
>> --
>> View this message in context:
>> http://old.nabble.com/Interaction-Between-Categorical-Variables_Another-Try--tp28076039p28100684.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
>>
>> =====================
>> 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
>
>

statisticsdoc

Re: Interaction Between Categorical Variables_Another Try?

Chao
Does the group mean stigma variable exhibit skew? Even when the distribution of individual scores is skewed, the distribution ofaggregate scores may be normally distributed.
HTH
Steve Brand
www.StatisticsDoc.com

-----Original Message-----
From: Chao Yawo <[hidden email]>
Date: Thu, 1 Apr 2010 14:38:39
To: <[hidden email]>
Subject: Re: Interaction Between Categorical Variables_Another Try?

Thanks --- Bruce, Ryan,and Steve,

Yes, I originally had a continuous variable that measures the
proportion of a respondent's community that hold certain stigmatizing
view (this was calculated by substracting the respondent's core from
the total score of all other respondents in the same stratum as the
respondent). I know an HLM model can handle this question better -
given the nested structure of the data. I plan to do it once I finish
going through some tutorials.

My expecting is extent of stigma (especially community level stigma)
will moderate the effect of risky behavior such as condom use on
desire to get HIV test. So, if stigma is high, then it would be less
likely for people to seek to test for HIV, even if they are engaged in
risky behaviors.

However, the continuous variable is a bit skewed, so I thought of
categorizing into three variables. And frankly, I was under the
impression that the categorical by categorical interaction was easier
to interpret than the categorical by continuous interaction effect
(but I may be wrong).

Thanks - Chao

==========
On Wed, Mar 31, 2010 at 7:43 PM, Statisticsdoc Consulting
<[hidden email]> wrote:

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