analyzing change for one dichotomous variable, assessed at pre and post

[pji]

I have a situation where I have one variable, smoking yes/no, taken at pre then taken at post, few years later. n ~70. same participants at both pre and post. the possible combinations are

smoking     pre       post
                 y          y
                 n          n
                 y          n
                 n          y

goal is to determine if the percentage of participants who changed from Y smoking at pre to N smoking at post is significant.

I don't think any sophisticated analysis has to be conducted. I would simply assign the four possible outcomes to four categories (y - y, n - n, y - n, n - y) and then run a chi-square to determine if the observed frequency counts of those four categories are different from the expected frequency counts. there is no post-hoc test, so I would have to heuristically examine which categories have the most deviance.

are there any analyses that could be conducted?
thank you in advance.
pj

[Bruce Weaver]

"goal is to determine if the percentage of participants who changed from Y smoking at pre to N smoking at post is significant."

Significantly different from what?  

The usual analysis for the situation you describe would be the McNemar chi-square test, aka the McNemar Change Test.  The data are in a 2x2 table with Yes & No for Pre in the rows, and Yes and No for post in the columns.  McNemar's chi-square is equivalent to a chi-square goodness of fit test on the two discordant cells,* with a null hypothesis specifying the same number of changes of each type (Y to N and N to Y).  (If the cell counts are too low, you can use a binomial test instead--SPSS computes one in that case, IIRC.)  Does that address the question you have in mind?  

* McNemar's test does not use the data in the concordant cells on the main diagonal.

pji wrote

I have a situation where I have one variable, smoking yes/no, taken at pre then taken at post, few years later. n ~70. same participants at both pre and post. the possible combinations are

smoking     pre       post
                 y          y
                 n          n
                 y          n
                 n          y

goal is to determine if the percentage of participants who changed from Y smoking at pre to N smoking at post is significant.

I don't think any sophisticated analysis has to be conducted. I would simply assign the four possible outcomes to four categories (y - y, n - n, y - n, n - y) and then run a chi-square to determine if the observed frequency counts of those four categories are different from the expected frequency counts. there is no post-hoc test, so I would have to heuristically examine which categories have the most deviance.

are there any analyses that could be conducted?
thank you in advance.
pj

[Hector Maletta]

On top of the very pertinent advice by Bruce, it should be useful to know
whether there was a 'treatment' or independent variable involved, and a
corresponding control group not subject to the treatment. What should be
significant is not the amount of change, but the difference in the
percentages changing among those undertaking the treatment and those in the
control group. In the absence of a control group, the percentage quitting
smoking may be (partly) an effect of some identifiable independent variable
(e.g. some exposure to advertising against smoking), and partly due to
random quitting that normally occurs among smokers. Since in the long term
people are quitting smoking (consistently reducing the percentage of people
that smoke), the percentage quitting might be statistically significant and
yet not be due to any particular intervention or treatment.
Also, the same 2x2 table tells not only about quitters, but also about the
number of people that did not smoke at the first occasion but became smokers
by the second time of observation. The flows in both directions may be
interesting (they may or may not offset each other). In this case, it would
be worthwhile assessing whether the difference between the two opposite
flows is or is not statistically significant.
As Bruce says, one should ask "significantly different from what".
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de Bruce
Weaver
Enviado el: Saturday, May 05, 2012 14:38
Para: [hidden email]
Asunto: Re: analyzing change for one dichotomous variable, assessed at pre
and post

"goal is to determine if the percentage of participants who changed from Y
smoking at pre to N smoking at post is significant."

Significantly different from what?

The usual analysis for the situation you describe would be the McNemar
chi-square test, aka the McNemar Change Test.  The data are in a 2x2 table
with Yes & No for Pre in the rows, and Yes and No for post in the columns.
McNemar's chi-square is equivalent to a chi-square goodness of fit test on
the two discordant cells,* with a null hypothesis specifying the same number
of changes of each type (Y to N and N to Y).  (If the cell counts are too
low, you can use a binomial test instead--SPSS computes one in that case,
IIRC.)  Does that address the question you have in mind?

* McNemar's test does not use the data in the concordant cells on the main
diagonal.


pji wrote

>
> I have a situation where I have one variable, smoking yes/no, taken at
> pre then taken at post, few years later. n ~70. same participants at
> both pre and post. the possible combinations are
>
> smoking     pre       post
>                  y          y
>                  n          n
>                  y          n
>                  n          y
>
> goal is to determine if the percentage of participants who changed
> from Y smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would
> simply assign the four possible outcomes to four categories (y - y, n
> - n, y - n, n - y) and then run a chi-square to determine if the
> observed frequency counts of those four categories are different from
> the expected frequency counts. there is no post-hoc test, so I would
> have to heuristically examine which categories have the most deviance.
>
> are there any analyses that could be conducted?
> thank you in advance.
> pj
>

-----
--
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://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dicho
tomous-variable-assessed-at-pre-and-post-tp5688136p5688233.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

[pji]

thank you both for your prompt responses.

 

there is no control or treatment group. the data were survey archival data.

 

good question, statistically different from the other conditions (Y - Y; N - N; N - Y). I agree that we want to determine if the percentage of those who are Y - N are different than the normal trajectory of quitting smoking that we would expect from the population. we don't have those population estimates, so we are just going with these sample estimates. so my inclination is to compare the percentage change of Y - N, to the remaining three outcomes.

 

the McNemar chi-square test does seem like the appropriate test and what I was originally thinking of, in addition to the chi-square analysis of the four outcome categories. it sounds like that should be sufficient.

 

comments? otherwise, thank you.

p

 

Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602

Office 16-403
312-662-4354
312-662-4099 (fax)
[hidden email]

---

From: Hector Maletta [via SPSSX Discussion] [[hidden email]]
Sent: Saturday, May 05, 2012 1:03 PM
To: Ji, Peter
Subject: Re: analyzing change for one dichotomous variable, assessed at pre and post

On top of the very pertinent advice by Bruce, it should be useful to know
whether there was a 'treatment' or independent variable involved, and a
corresponding control group not subject to the treatment. What should be
significant is not the amount of change, but the difference in the
percentages changing among those undertaking the treatment and those in the
control group. In the absence of a control group, the percentage quitting
smoking may be (partly) an effect of some identifiable independent variable
(e.g. some exposure to advertising against smoking), and partly due to
random quitting that normally occurs among smokers. Since in the long term
people are quitting smoking (consistently reducing the percentage of people
that smoke), the percentage quitting might be statistically significant and
yet not be due to any particular intervention or treatment.
Also, the same 2x2 table tells not only about quitters, but also about the
number of people that did not smoke at the first occasion but became smokers
by the second time of observation. The flows in both directions may be
interesting (they may or may not offset each other). In this case, it would
be worthwhile assessing whether the difference between the two opposite
flows is or is not statistically significant.
As Bruce says, one should ask "significantly different from what".
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688262&amp;i=0" rel=nofollow target=_blank>[hidden email]] En nombre de Bruce
Weaver
Enviado el: Saturday, May 05, 2012 14:38
Para: <A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688262&amp;i=1" rel=nofollow target=_blank>[hidden email]
Asunto: Re: analyzing change for one dichotomous variable, assessed at pre
and post

"goal is to determine if the percentage of participants who changed from Y
smoking at pre to N smoking at post is significant."

Significantly different from what?

The usual analysis for the situation you describe would be the McNemar
chi-square test, aka the McNemar Change Test.  The data are in a 2x2 table
with Yes & No for Pre in the rows, and Yes and No for post in the columns.
McNemar's chi-square is equivalent to a chi-square goodness of fit test on
the two discordant cells,* with a null hypothesis specifying the same number
of changes of each type (Y to N and N to Y).  (If the cell counts are too
low, you can use a binomial test instead--SPSS computes one in that case,
IIRC.)  Does that address the question you have in mind?

* McNemar's test does not use the data in the concordant cells on the main
diagonal.


pji wrote

>
> I have a situation where I have one variable, smoking yes/no, taken at
> pre then taken at post, few years later. n ~70. same participants at
> both pre and post. the possible combinations are
>
> smoking     pre       post
>                  y          y
>                  n          n
>                  y          n
>                  n          y
>
> goal is to determine if the percentage of participants who changed
> from Y smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would
> simply assign the four possible outcomes to four categories (y - y, n
> - n, y - n, n - y) and then run a chi-square to determine if the
> observed frequency counts of those four categories are different from
> the expected frequency counts. there is no post-hoc test, so I would
> have to heuristically examine which categories have the most deviance.
>
> are there any analyses that could be conducted?
> thank you in advance.
> pj
>

-----
--
Bruce Weaver
<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688262&amp;i=2" rel=nofollow target=_blank>[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://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dicho
tomous-variable-assessed-at-pre-and-post-tp5688136p5688233.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688262&amp;i=3" rel=nofollow target=_blank>[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
<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688262&amp;i=4" rel=nofollow target=_blank>[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

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688262.html

To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here.
NAML

[Bruce Weaver]

It's still not clear to me how you intend to generate the expected counts for the analysis that includes all 4 cells.  Can you clarify that?

Thanks.

pji wrote

thank you both for your prompt responses.

there is no control or treatment group. the data were survey archival data.

good question, statistically different from the other conditions (Y - Y; N - N; N - Y). I agree that we want to determine if the percentage of those who are Y - N are different than the normal trajectory of quitting smoking that we would expect from the population. we don't have those population estimates, so we are just going with these sample estimates. so my inclination is to compare the percentage change of Y - N, to the remaining three outcomes.

the McNemar chi-square test does seem like the appropriate test and what I was originally thinking of, in addition to the chi-square analysis of the four outcome categories. it sounds like that should be sufficient.

comments? otherwise, thank you.
p

Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602
Office 16-403
312-662-4354
312-662-4099 (fax)
[hidden email]
________________________________
From: Hector Maletta [via SPSSX Discussion] [[hidden email]]
Sent: Saturday, May 05, 2012 1:03 PM
To: Ji, Peter
Subject: Re: analyzing change for one dichotomous variable, assessed at pre and post

On top of the very pertinent advice by Bruce, it should be useful to know
whether there was a 'treatment' or independent variable involved, and a
corresponding control group not subject to the treatment. What should be
significant is not the amount of change, but the difference in the
percentages changing among those undertaking the treatment and those in the
control group. In the absence of a control group, the percentage quitting
smoking may be (partly) an effect of some identifiable independent variable
(e.g. some exposure to advertising against smoking), and partly due to
random quitting that normally occurs among smokers. Since in the long term
people are quitting smoking (consistently reducing the percentage of people
that smoke), the percentage quitting might be statistically significant and
yet not be due to any particular intervention or treatment.
Also, the same 2x2 table tells not only about quitters, but also about the
number of people that did not smoke at the first occasion but became smokers
by the second time of observation. The flows in both directions may be
interesting (they may or may not offset each other). In this case, it would
be worthwhile assessing whether the difference between the two opposite
flows is or is not statistically significant.
As Bruce says, one should ask "significantly different from what".
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=0>] En nombre de Bruce
Weaver
Enviado el: Saturday, May 05, 2012 14:38
Para: [hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=1>
Asunto: Re: analyzing change for one dichotomous variable, assessed at pre
and post

"goal is to determine if the percentage of participants who changed from Y
smoking at pre to N smoking at post is significant."

Significantly different from what?

The usual analysis for the situation you describe would be the McNemar
chi-square test, aka the McNemar Change Test.  The data are in a 2x2 table
with Yes & No for Pre in the rows, and Yes and No for post in the columns.
McNemar's chi-square is equivalent to a chi-square goodness of fit test on
the two discordant cells,* with a null hypothesis specifying the same number
of changes of each type (Y to N and N to Y).  (If the cell counts are too
low, you can use a binomial test instead--SPSS computes one in that case,
IIRC.)  Does that address the question you have in mind?

* McNemar's test does not use the data in the concordant cells on the main
diagonal.


pji wrote

>
> I have a situation where I have one variable, smoking yes/no, taken at
> pre then taken at post, few years later. n ~70. same participants at
> both pre and post. the possible combinations are
>
> smoking     pre       post
>                  y          y
>                  n          n
>                  y          n
>                  n          y
>
> goal is to determine if the percentage of participants who changed
> from Y smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would
> simply assign the four possible outcomes to four categories (y - y, n
> - n, y - n, n - y) and then run a chi-square to determine if the
> observed frequency counts of those four categories are different from
> the expected frequency counts. there is no post-hoc test, so I would
> have to heuristically examine which categories have the most deviance.
>
> are there any analyses that could be conducted?
> thank you in advance.
> pj
>


-----
--
Bruce Weaver
[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=2>
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://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dicho
tomous-variable-assessed-at-pre-and-post-tp5688136p5688233.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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=3> (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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=4> (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


________________________________
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688262.html
To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=unsubscribe_by_code&node=5688136&code=cGppQGFkbGVyLmVkdXw1Njg4MTM2fDg0MTc5OTk=>.
NAML<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>

[pji]

this is my thought.

 

 

 

category 1, n = 6; category 2, n = 3; category 3, n = 5; category 4, n = 1.

just as an example.

 

hope that helps, thanks.

p

 

 

          

Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602

Office 16-403
312-662-4354
312-662-4099 (fax)
[hidden email]

---

From: Bruce Weaver [via SPSSX Discussion] [[hidden email]]
Sent: Saturday, May 05, 2012 3:30 PM
To: Ji, Peter
Subject: RE: analyzing change for one dichotomous variable, assessed at pre and post

It's still not clear to me how you intend to generate the expected counts for the analysis that includes all 4 cells.  Can you clarify that?

Thanks.

pji wrote

thank you both for your prompt responses.

there is no control or treatment group. the data were survey archival data.

good question, statistically different from the other conditions (Y - Y; N - N; N - Y). I agree that we want to determine if the percentage of those who are Y - N are different than the normal trajectory of quitting smoking that we would expect from the population. we don't have those population estimates, so we are just going with these sample estimates. so my inclination is to compare the percentage change of Y - N, to the remaining three outcomes.

the McNemar chi-square test does seem like the appropriate test and what I was originally thinking of, in addition to the chi-square analysis of the four outcome categories. it sounds like that should be sufficient.

comments? otherwise, thank you.
p

Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602
Office 16-403
312-662-4354
312-662-4099 (fax)
<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688453&amp;i=0" rel=nofollow target=_blank>[hidden email]
________________________________
From: Hector Maletta [via SPSSX Discussion] [<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688453&amp;i=1" rel=nofollow target=_blank>[hidden email]]
Sent: Saturday, May 05, 2012 1:03 PM
To: Ji, Peter
Subject: Re: analyzing change for one dichotomous variable, assessed at pre and post

On top of the very pertinent advice by Bruce, it should be useful to know
whether there was a 'treatment' or independent variable involved, and a
corresponding control group not subject to the treatment. What should be
significant is not the amount of change, but the difference in the
percentages changing among those undertaking the treatment and those in the
control group. In the absence of a control group, the percentage quitting
smoking may be (partly) an effect of some identifiable independent variable
(e.g. some exposure to advertising against smoking), and partly due to
random quitting that normally occurs among smokers. Since in the long term
people are quitting smoking (consistently reducing the percentage of people
that smoke), the percentage quitting might be statistically significant and
yet not be due to any particular intervention or treatment.
Also, the same 2x2 table tells not only about quitters, but also about the
number of people that did not smoke at the first occasion but became smokers
by the second time of observation. The flows in both directions may be
interesting (they may or may not offset each other). In this case, it would
be worthwhile assessing whether the difference between the two opposite
flows is or is not statistically significant.
As Bruce says, one should ask "significantly different from what".
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=0>] En nombre de Bruce
Weaver
Enviado el: Saturday, May 05, 2012 14:38
Para: [hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=1>
Asunto: Re: analyzing change for one dichotomous variable, assessed at pre
and post

"goal is to determine if the percentage of participants who changed from Y
smoking at pre to N smoking at post is significant."

Significantly different from what?

The usual analysis for the situation you describe would be the McNemar
chi-square test, aka the McNemar Change Test.  The data are in a 2x2 table
with Yes & No for Pre in the rows, and Yes and No for post in the columns.
McNemar's chi-square is equivalent to a chi-square goodness of fit test on
the two discordant cells,* with a null hypothesis specifying the same number
of changes of each type (Y to N and N to Y).  (If the cell counts are too
low, you can use a binomial test instead--SPSS computes one in that case,
IIRC.)  Does that address the question you have in mind?

* McNemar's test does not use the data in the concordant cells on the main
diagonal.


pji wrote

>
> I have a situation where I have one variable, smoking yes/no, taken at
> pre then taken at post, few years later. n ~70. same participants at
> both pre and post. the possible combinations are
>
> smoking     pre       post
>                  y          y
>                  n          n
>                  y          n
>                  n          y
>
> goal is to determine if the percentage of participants who changed
> from Y smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would
> simply assign the four possible outcomes to four categories (y - y, n
> - n, y - n, n - y) and then run a chi-square to determine if the
> observed frequency counts of those four categories are different from
> the expected frequency counts. there is no post-hoc test, so I would
> have to heuristically examine which categories have the most deviance.
>
> are there any analyses that could be conducted?
> thank you in advance.
> pj
>

-----
--
Bruce Weaver
[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=2>
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://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dicho
tomous-variable-assessed-at-pre-and-post-tp5688136p5688233.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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=3> (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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=4> (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


________________________________
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688262.html
To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here<<A href="" rel=nofollow target=_blank>
NAML<<A href="http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&amp;id=instant_html%21nabble%3Aemail.naml&amp;base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&amp;breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>" rel=nofollow target=_blank>http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688453.html

To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here.
NAML

[Bruce Weaver]

"category 1, n = 6; category 2, n = 3; category 3, n = 5; category 4, n = 1."

Those are observed counts.  What are the expected counts for that example under your null hypothesis?

pji wrote

this is my thought.


[cid:107FE68-B6C4-41077-A13C-B60E38FAB6@MimeCtl]

category 1, n = 6; category 2, n = 3; category 3, n = 5; category 4, n = 1.
just as an example.

hope that helps, thanks.
p



Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602
Office 16-403
312-662-4354
312-662-4099 (fax)
[hidden email]
________________________________
From: Bruce Weaver [via SPSSX Discussion] [[hidden email]]
Sent: Saturday, May 05, 2012 3:30 PM
To: Ji, Peter
Subject: RE: analyzing change for one dichotomous variable, assessed at pre and post

It's still not clear to me how you intend to generate the expected counts for the analysis that includes all 4 cells.  Can you clarify that?

Thanks.


pji wrote
thank you both for your prompt responses.

there is no control or treatment group. the data were survey archival data.

good question, statistically different from the other conditions (Y - Y; N - N; N - Y). I agree that we want to determine if the percentage of those who are Y - N are different than the normal trajectory of quitting smoking that we would expect from the population. we don't have those population estimates, so we are just going with these sample estimates. so my inclination is to compare the percentage change of Y - N, to the remaining three outcomes.

the McNemar chi-square test does seem like the appropriate test and what I was originally thinking of, in addition to the chi-square analysis of the four outcome categories. it sounds like that should be sufficient.

comments? otherwise, thank you.
p

Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602
Office 16-403
312-662-4354
312-662-4099 (fax)
[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688453&i=0>
________________________________
From: Hector Maletta [via SPSSX Discussion] [[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688453&i=1>]
Sent: Saturday, May 05, 2012 1:03 PM
To: Ji, Peter
Subject: Re: analyzing change for one dichotomous variable, assessed at pre and post

On top of the very pertinent advice by Bruce, it should be useful to know
whether there was a 'treatment' or independent variable involved, and a
corresponding control group not subject to the treatment. What should be
significant is not the amount of change, but the difference in the
percentages changing among those undertaking the treatment and those in the
control group. In the absence of a control group, the percentage quitting
smoking may be (partly) an effect of some identifiable independent variable
(e.g. some exposure to advertising against smoking), and partly due to
random quitting that normally occurs among smokers. Since in the long term
people are quitting smoking (consistently reducing the percentage of people
that smoke), the percentage quitting might be statistically significant and
yet not be due to any particular intervention or treatment.
Also, the same 2x2 table tells not only about quitters, but also about the
number of people that did not smoke at the first occasion but became smokers
by the second time of observation. The flows in both directions may be
interesting (they may or may not offset each other). In this case, it would
be worthwhile assessing whether the difference between the two opposite
flows is or is not statistically significant.
As Bruce says, one should ask "significantly different from what".
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=0>] En nombre de Bruce
Weaver
Enviado el: Saturday, May 05, 2012 14:38
Para: [hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=1>
Asunto: Re: analyzing change for one dichotomous variable, assessed at pre
and post

"goal is to determine if the percentage of participants who changed from Y
smoking at pre to N smoking at post is significant."

Significantly different from what?

The usual analysis for the situation you describe would be the McNemar
chi-square test, aka the McNemar Change Test.  The data are in a 2x2 table
with Yes & No for Pre in the rows, and Yes and No for post in the columns.
McNemar's chi-square is equivalent to a chi-square goodness of fit test on
the two discordant cells,* with a null hypothesis specifying the same number
of changes of each type (Y to N and N to Y).  (If the cell counts are too
low, you can use a binomial test instead--SPSS computes one in that case,
IIRC.)  Does that address the question you have in mind?

* McNemar's test does not use the data in the concordant cells on the main
diagonal.


pji wrote

>
> I have a situation where I have one variable, smoking yes/no, taken at
> pre then taken at post, few years later. n ~70. same participants at
> both pre and post. the possible combinations are
>
> smoking     pre       post
>                  y          y
>                  n          n
>                  y          n
>                  n          y
>
> goal is to determine if the percentage of participants who changed
> from Y smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would
> simply assign the four possible outcomes to four categories (y - y, n
> - n, y - n, n - y) and then run a chi-square to determine if the
> observed frequency counts of those four categories are different from
> the expected frequency counts. there is no post-hoc test, so I would
> have to heuristically examine which categories have the most deviance.
>
> are there any analyses that could be conducted?
> thank you in advance.
> pj
>


-----
--
Bruce Weaver
[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=2>
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://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dicho
tomous-variable-assessed-at-pre-and-post-tp5688136p5688233.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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=3> (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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=4> (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


________________________________
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688262.html
To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=unsubscribe_by_code&node=5688136&code=cGppQGFkbGVyLmVkdXw1Njg4MTM2fDg0MTc5OTk=>.
NAML<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>
--
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.


________________________________
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688453.html
To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=unsubscribe_by_code&node=5688136&code=cGppQGFkbGVyLmVkdXw1Njg4MTM2fDg0MTc5OTk=>.
NAML<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>

[Kornbrot, Diana]

classic solution is mcnemar test
test
y-n/(y-n + n-y) > .5
use exact binomial if n switchers small people wio do not chane irrelevant
best
diana

Sent from my iPhone

On 6 May 2012, at 05:36 AM, "pji" <[hidden email]> wrote:

> I have a situation where I have one variable, smoking yes/no, taken at pre
> then taken at post, few years later. n ~70. same participants at both pre
> and post. the possible combinations are
>
> smoking     pre       post
>                 y          y
>                 n          n
>                 y          n
>                 n          y
>
> goal is to determine if the percentage of participants who changed from Y
> smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would simply
> assign the four possible outcomes to four categories (y - y, n - n, y - n, n
> - y) and then run a chi-square to determine if the observed frequency counts
> of those four categories are different from the expected frequency counts.
> there is no post-hoc test, so I would have to heuristically examine which
> categories have the most deviance.
>
> are there any analyses that could be conducted?
> thank you in advance.
> pj
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136.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

[pji]

Thanks Bruce, sorry I misinterpreted.

as you might expect, there is no apriori expected counts for the null, other than the expected counts would follow the proportions of 25% across all categories, which is probably not a reasonable expectation, but we don't have an estimate. with that in mind, I'm guessing that the 4 category analysis I proposed below would not be appropriate because of the lack of an expected count hypothesis.

thanks for these comments.

p

 

 

 

---

From: Bruce Weaver [via SPSSX Discussion] [[hidden email]]
Sent: Saturday, May 05, 2012 6:26 PM
To: Ji, Peter
Subject: RE: analyzing change for one dichotomous variable, assessed at pre and post

"category 1, n = 6; category 2, n = 3; category 3, n = 5; category 4, n = 1."

Those are observed counts.  What are the expected counts for that example under your null hypothesis?

pji wrote

this is my thought.


[cid:107FE68-B6C4-41077-A13C-B60E38FAB6@MimeCtl]

category 1, n = 6; category 2, n = 3; category 3, n = 5; category 4, n = 1.
just as an example.

hope that helps, thanks.
p



Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602
Office 16-403
312-662-4354
312-662-4099 (fax)
<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688623&amp;i=0" rel=nofollow target=_blank>[hidden email]
________________________________
From: Bruce Weaver [via SPSSX Discussion] [<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688623&amp;i=1" rel=nofollow target=_blank>[hidden email]]
Sent: Saturday, May 05, 2012 3:30 PM
To: Ji, Peter
Subject: RE: analyzing change for one dichotomous variable, assessed at pre and post

It's still not clear to me how you intend to generate the expected counts for the analysis that includes all 4 cells.  Can you clarify that?

Thanks.


pji wrote
thank you both for your prompt responses.

there is no control or treatment group. the data were survey archival data.

good question, statistically different from the other conditions (Y - Y; N - N; N - Y). I agree that we want to determine if the percentage of those who are Y - N are different than the normal trajectory of quitting smoking that we would expect from the population. we don't have those population estimates, so we are just going with these sample estimates. so my inclination is to compare the percentage change of Y - N, to the remaining three outcomes.

the McNemar chi-square test does seem like the appropriate test and what I was originally thinking of, in addition to the chi-square analysis of the four outcome categories. it sounds like that should be sufficient.

comments? otherwise, thank you.
p

Peter Ji, Ph.D.
Core Faculty
Adler School of Professional Psychology
17 North Dearborn Street
Chicago, IL 60602
Office 16-403
312-662-4354
312-662-4099 (fax)
[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688453&i=0>
________________________________
From: Hector Maletta [via SPSSX Discussion] [[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688453&i=1>]
Sent: Saturday, May 05, 2012 1:03 PM
To: Ji, Peter
Subject: Re: analyzing change for one dichotomous variable, assessed at pre and post

On top of the very pertinent advice by Bruce, it should be useful to know
whether there was a 'treatment' or independent variable involved, and a
corresponding control group not subject to the treatment. What should be
significant is not the amount of change, but the difference in the
percentages changing among those undertaking the treatment and those in the
control group. In the absence of a control group, the percentage quitting
smoking may be (partly) an effect of some identifiable independent variable
(e.g. some exposure to advertising against smoking), and partly due to
random quitting that normally occurs among smokers. Since in the long term
people are quitting smoking (consistently reducing the percentage of people
that smoke), the percentage quitting might be statistically significant and
yet not be due to any particular intervention or treatment.
Also, the same 2x2 table tells not only about quitters, but also about the
number of people that did not smoke at the first occasion but became smokers
by the second time of observation. The flows in both directions may be
interesting (they may or may not offset each other). In this case, it would
be worthwhile assessing whether the difference between the two opposite
flows is or is not statistically significant.
As Bruce says, one should ask "significantly different from what".
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=0>] En nombre de Bruce
Weaver
Enviado el: Saturday, May 05, 2012 14:38
Para: [hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=1>
Asunto: Re: analyzing change for one dichotomous variable, assessed at pre
and post

"goal is to determine if the percentage of participants who changed from Y
smoking at pre to N smoking at post is significant."

Significantly different from what?

The usual analysis for the situation you describe would be the McNemar
chi-square test, aka the McNemar Change Test.  The data are in a 2x2 table
with Yes & No for Pre in the rows, and Yes and No for post in the columns.
McNemar's chi-square is equivalent to a chi-square goodness of fit test on
the two discordant cells,* with a null hypothesis specifying the same number
of changes of each type (Y to N and N to Y).  (If the cell counts are too
low, you can use a binomial test instead--SPSS computes one in that case,
IIRC.)  Does that address the question you have in mind?

* McNemar's test does not use the data in the concordant cells on the main
diagonal.


pji wrote

>
> I have a situation where I have one variable, smoking yes/no, taken at
> pre then taken at post, few years later. n ~70. same participants at
> both pre and post. the possible combinations are
>
> smoking     pre       post
>                  y          y
>                  n          n
>                  y          n
>                  n          y
>
> goal is to determine if the percentage of participants who changed
> from Y smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would
> simply assign the four possible outcomes to four categories (y - y, n
> - n, y - n, n - y) and then run a chi-square to determine if the
> observed frequency counts of those four categories are different from
> the expected frequency counts. there is no post-hoc test, so I would
> have to heuristically examine which categories have the most deviance.
>
> are there any analyses that could be conducted?
> thank you in advance.
> pj
>

-----
--
Bruce Weaver
[hidden email]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=2>
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://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dicho
tomous-variable-assessed-at-pre-and-post-tp5688136p5688233.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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=3> (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]<thismessage:/user/SendEmail.jtp?type=node&node=5688262&i=4> (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


________________________________
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688262.html
To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here<<A href="" rel=nofollow target=_blank>
NAML<<A href="http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&amp;id=instant_html%21nabble%3Aemail.naml&amp;base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&amp;breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>" rel=nofollow target=_blank>http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>
--
Bruce Weaver
<A href="thismessage:/user/SendEmail.jtp?type=node&amp;node=5688623&amp;i=2" rel=nofollow target=_blank>[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.


________________________________
If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688453.html
To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here<<A href="" rel=nofollow target=_blank>
NAML<<A href="http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&amp;id=instant_html%21nabble%3Aemail.naml&amp;base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&amp;breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>" rel=nofollow target=_blank>http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/analyzing-change-for-one-dichotomous-variable-assessed-at-pre-and-post-tp5688136p5688623.html

To unsubscribe from analyzing change for one dichotomous variable, assessed at pre and post, click here.
NAML

[Rich Ulrich]

Create a test using these data alone? (No rates available
as norms.)

No, there is no interesting test available.

If you had half the smokers quit, that would seem to me to
be a "large number" -- depending on ages and years of gap --
but there is nothing available to say that it is "unexpected"
and should be anything other than a curious sample/ population
statistic.

And I would see no reason to expect the number quitting
to match the number starting, which is the null hypothesis
of McNemar's test.

--
Rich Ulrich

> Date: Sat, 5 May 2012 09:21:17 -0700

> From: [hidden email]
> Subject: analyzing change for one dichotomous variable, assessed at pre and post
> To: [hidden email]
>
> I have a situation where I have one variable, smoking yes/no, taken at pre
> then taken at post, few years later. n ~70. same participants at both pre
> and post. the possible combinations are
>
> smoking pre post
> y y
> n n
> y n
> n y
>
> goal is to determine if the percentage of participants who changed from Y
> smoking at pre to N smoking at post is significant.
>
> I don't think any sophisticated analysis has to be conducted. I would simply
> assign the four possible outcomes to four categories (y - y, n - n, y - n, n
> - y) and then run a chi-square to determine if the observed frequency counts
> of those four categories are different from the expected frequency counts.
> there is no post-hoc test, so I would have to heuristically examine which
> categories have the most deviance.
>
> are there any analyses that could be conducted?

...