T-tests - but what else

[poloboyden]

Hello
I have a database of data with two groups: psychiatric clients, and depressed controls.

To explore the differences between them I am using independent samples t-tests. this shows a sig diffs between the two groups on my main measure.

However, I also have other continuous variables that may predict or affect or account for the significant results (e.g. IQ, anxiety, depression).

Should I just do correlations and t-tests or can I do an analysis of variance at all (but I only have two independent groups!)

Thanks!!!!!

[Bruce Weaver]

If you run a linear regression model with GROUP as the lone explanatory variable and your main measure as the dependent variable, you should find that you get results equivalent to the independent groups t-test.  (It's convenient to have 0-1 Group coding when doing this, so that the intercept gives the mean for Group 0, and the coefficient for Group gives the mean difference between the groups.)  Having done that, you can run another model that also includes the other continuous explanatory variables, thereby "controlling" for them when examining the difference between groups.

However...what are the sample sizes for your two groups?  If you don't have enough data, you could overfit the model by including too many variables.

HTH.

poloboyden wrote

Hello
I have a database of data with two groups: psychiatric clients, and depressed controls.

To explore the differences between them I am using independent samples t-tests. this shows a sig diffs between the two groups on my main measure.

However, I also have other continuous variables that may predict or affect or account for the significant results (e.g. IQ, anxiety, depression).

Should I just do correlations and t-tests or can I do an analysis of variance at all (but I only have two independent groups!)

Thanks!!!!!

[poloboyden]

Thank you so much Bruce

The sample size is 34 with 17 people in both groups. So not very big at all. There's no sig diff between the groups on age IQ gender because they are matched. There is a sig diff for depression though.

So with such a small sample size is it worth doing a regression?

Sent from my iPhone

On 10 May 2012, at 22:13, "Bruce Weaver [via SPSSX Discussion]" <[hidden email]> wrote:

If you run a linear regression model with GROUP as the lone explanatory variable and your main measure as the dependent variable, you should find that you get results equivalent to the independent groups t-test.  (It's convenient to have 0-1 Group coding when doing this, so that the intercept gives the mean for Group 0, and the coefficient for Group gives the mean difference between the groups.)  Having done that, you can run another model that also includes the other continuous explanatory variables, thereby "controlling" for them when examining the difference between groups.

However...what are the sample sizes for your two groups?  If you don't have enough data, you could overfit the model by including too many variables.

HTH.

poloboyden wrote

Hello
I have a database of data with two groups: psychiatric clients, and depressed controls.

To explore the differences between them I am using independent samples t-tests. this shows a sig diffs between the two groups on my main measure.

However, I also have other continuous variables that may predict or affect or account for the significant results (e.g. IQ, anxiety, depression).

Should I just do correlations and t-tests or can I do an analysis of variance at all (but I only have two independent groups!)

Thanks!!!!!

---

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

http://spssx-discussion.1045642.n5.nabble.com/T-tests-but-what-else-tp5701479p5701506.html

To unsubscribe from T-tests - but what else, click here.
NAML

[Poes, Matthew Joseph]

I would argue with that sample size, you should do nothing more than the T-test, and that you should carefully examine the distribution of the variables, and check all assumptions.  Many people will argue you want at least 30 people in each group, or if less, at least 10 as long as all assumptions are well met.  I would also argue with such a small sample size that you might want to consider a method which uses a resampling technique and compare group means by confidence intervals. 

 

If you used bootstrapping, while less than ideal, you might get away with conditional model that contains a covariate, but I would not go beyond one covariate, and again, check all assumptions carefully.  I would also cautiously interpret that model, and any write-up would need to treat it as a secondary analysis to help inform future research, not as a primary finding.

 

Matthew J Poes

Research Data Specialist

Center for Prevention Research and Development

University of Illinois

510 Devonshire Dr.

Champaign, IL 61820

Phone: [phone]

email: [hidden email]

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of poloboyden
Sent: Thursday, May 10, 2012 4:35 PM
To: [hidden email]
Subject: Re: T-tests - but what else

 

Thank you so much Bruce

 

The sample size is 34 with 17 people in both groups. So not very big at all. There's no sig diff between the groups on age IQ gender because they are matched. There is a sig diff for depression though.

 

So with such a small sample size is it worth doing a regression?

Sent from my iPhone

On 10 May 2012, at 22:13, "Bruce Weaver [via SPSSX Discussion]" <[hidden email]> wrote:

If you run a linear regression model with GROUP as the lone explanatory variable and your main measure as the dependent variable, you should find that you get results equivalent to the independent groups t-test.  (It's convenient to have 0-1 Group coding when doing this, so that the intercept gives the mean for Group 0, and the coefficient for Group gives the mean difference between the groups.)  Having done that, you can run another model that also includes the other continuous explanatory variables, thereby "controlling" for them when examining the difference between groups.

However...what are the sample sizes for your two groups?  If you don't have enough data, you could overfit the model by including too many variables.

HTH.

poloboyden wrote

Hello
I have a database of data with two groups: psychiatric clients, and depressed controls.

To explore the differences between them I am using independent samples t-tests. this shows a sig diffs between the two groups on my main measure.

However, I also have other continuous variables that may predict or affect or account for the significant results (e.g. IQ, anxiety, depression).

Should I just do correlations and t-tests or can I do an analysis of variance at all (but I only have two independent groups!)

Thanks!!!!!

--
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/T-tests-but-what-else-tp5701479p5701506.html

To unsubscribe from T-tests - but what else, click here.
NAML

 

---

View this message in context: Re: T-tests - but what else
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

[Harmon, Judith, PED]

I will be out of the office on from Friday May 10th through Monday May 14th. If you have questions about the ADE system, please call Craig at 1-800-334-1918.   Any other questions, please contact Dan Hall, at 331-0415.  Thanks!

[Jenny Bartashnik]

Hi,

 

Thank you for contacting Reputation Institute’s MSG.

 

I’ll be out of the office Fri May 11 and will be back on Mon May 14.

 

If you need immediate assistance please contact me on my cell: 917-415-2843

 

Best,

Jenny

 

[Art Kendall]

How did you select these cases?  What is the distinction between psychiatric clients and depressed _comparison_ people?

How did you do the matching?



On 5/10/2012 5:34 PM, poloboyden wrote:

Thank you so much Bruce

The sample size is 34 with 17 people in both groups. So not very big at all. There's no sig diff between the groups on age IQ gender because they are matched. There is a sig diff for depression though.

So with such a small sample size is it worth doing a regression?

Sent from my iPhone

On 10 May 2012, at 22:13, "Bruce Weaver [via SPSSX Discussion]" <[hidden email]> wrote:

If you run a linear regression model with GROUP as the lone explanatory variable and your main measure as the dependent variable, you should find that you get results equivalent to the independent groups t-test.  (It's convenient to have 0-1 Group coding when doing this, so that the intercept gives the mean for Group 0, and the coefficient for Group gives the mean difference between the groups.)  Having done that, you can run another model that also includes the other continuous explanatory variables, thereby "controlling" for them when examining the difference between groups.

However...what are the sample sizes for your two groups?  If you don't have enough data, you could overfit the model by including too many variables.

HTH.

poloboyden wrote

Hello
I have a database of data with two groups: psychiatric clients, and depressed controls.

To explore the differences between them I am using independent samples t-tests. this shows a sig diffs between the two groups on my main measure.

However, I also have other continuous variables that may predict or affect or account for the significant results (e.g. IQ, anxiety, depression).

Should I just do correlations and t-tests or can I do an analysis of variance at all (but I only have two independent groups!)

Thanks!!!!!

---

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

http://spssx-discussion.1045642.n5.nabble.com/T-tests-but-what-else-tp5701479p5701506.html

To unsubscribe from T-tests - but what else, click here.
NAML

---

View this message in context: Re: T-tests - but what else
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

[poloboyden]

I selected them on the basis of being diagnosed schizophrenic (psychiatric group) or depressed (control group) They volunteered to take part.

I matched simply by recruiting several participants and comparing means along the way to ensure they were similar.

[Art Kendall]

But did you match pairs of cases so that each pair had the same age gender and IQ?

Did you have recruits for each group for whom you did not find a match?

What I am getting at is that you may have more cases to work with and/or that you might have some reduction in error (residual) terms due to matching.

On 5/13/2012 8:56 AM, poloboyden wrote:

I selected them on the basis of being diagnosed schizophrenic (psychiatric
group) or depressed (control group) They volunteered to take part.

I matched simply by recruiting several participants and comparing means
along the way to ensure they were similar.


--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/T-tests-but-what-else-tp5701479p5708156.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

[poloboyden]

I see - but no, this was not a matched pairs design. Rather, it is a case-control design, but I have tried my best to ensure both groups are not significantly different for age, gender and IQ (and they aren't).

I am simply exploring the hypotheses in a preliminary study exploring whether there are differences between groups on two measures.

The measures are normally distributed, which has led me to then assume I can use an independent samples t-test (which are significant).

I have about 9 other measures/variables too, of which some show sig differences between the two groups too (e.g. depression, mania).

So, I think it looks like t-tests and correlations are the only things I can (and perhaps should) use considering it wasnt matched pairs, and the sample is only 34 (n=17 in both groups)?

[Rich Ulrich]

Even if you had collected it as "matched pairs", the more efficient
and more robust analysis - compared to a paired t-test -
ordinarily would be the test from the ANCOVA, where sex,
age and IQ are used as covariates.

If you are considering symptoms like depression, you might
also consider entering the largest of these as another covariate,
to demonstrate whether or not any of the others can be held to
have an independent effect, beyond their correlation with depression.

With a sample this small, it is quite possible that the largest one
will wipe out the "significance" of the rest.  And that can simplify
the picture that you present.   It is less interesting to see whether
two symptoms wipe out the rest (if one does not), because using
two is likely to have that result, regardless of which two you use.

--
Rich Ulrich

> Date: Sun, 13 May 2012 11:02:49 -0700

> From: [hidden email]
> Subject: Re: T-tests - but what else
> To: [hidden email]
>
> I see - but no, this was not a matched pairs design. Rather, it is a
> case-control design, but I have tried my best to ensure both groups are not
> significantly different for age, gender and IQ (and they aren't).
>
> I am simply exploring the hypotheses in a preliminary study exploring
> whether there are differences between groups on two measures.,
>
> The measures are normally distributed, which has led me to then assume I can
> use an independent samples t-test (which are significant).
>
> I have about 9 other measures/variables too, of which some show sig
> differences between the two groups too (e.g. depression, mania).
>
> So, I think it looks like t-tests and correlations are the only things I can
> (and perhaps should) use considering it wasnt matched pairs, and the sample
> is only 34 (n=17 in both groups)?
>
> --