Confusion with Independent Samples T Test

[holly costar]

Hi,

I'm a graduate student working with the SPSS graduate pack and keep getting
warnings when trying to execute an independent samples t test. The warnings
say that one of the groups is empty and the test can't be executed. I've
defined the groups (using a string variable as grouping, y/n coded 1/2 for
whether or not participants have health insurance) and there are no gaps in
the data. Any idea what I'm doing wrong?

Help would be much appreciated!

thanks,
Holly

[Richard Ristow]

At 10:31 AM 6/29/2007, holly costar wrote:

>I'm getting warnings when trying to execute an independent samples t
>test. The warnings say that one of the groups is empty and the test
>can't be executed. I've defined the groups (using a string variable as
>grouping, y/n coded 1/2 for whether or not participants have health
>insurance) and there are no gaps in the data. Any idea what I'm doing
>wrong?

No idea offhand, but here are some quick diagnostic tools:

- Run FREQUENCIES on your grouping variable. If there are no counts in
one category, that's the problem. Then, it may be,
. You defined the grouping variable using SPSS transformations, and
there's a bug in your logic - so, check the logic
. You don't have the population you thought you had (sorry, but this
does happen). To check that, run descriptive statistics that will
characterize your data relative to the population you expected
. If you do see members in both groups, perhaps you specified the
categories wrong on the /GROUPS subcommand

Forgive me if these sound too simple-minded. The fact is, however
experienced we are, a lot of our mistakes are something very simple.

- Run MEANS, for your independent variable by your grouping variable.
Of course, if T-TEST finds a category empty, so will MEANS. But
. If FREQUENCIES shows members in a group and MEANS doesn't, then all
values of the dependent are missing in that group. That can come from
many sources, including logic errors and oddities in how the data was
gathered in the first place. If there are few valid values for any
group, your dependent variable is sparse, arising either from a mistake
in data-gathering or computation; or data that really *is* sparse,
which usually takes a little rethinking of the study.

Wild shots, of course. But these are what I'd do first, given your
situation.

-Best of luck,
  Richard Ristow

[Oliver, Richard]

The following sounds suspect to me:

"I've defined the groups (using a string variable as grouping, y/n coded 1/2 ..."

Is the grouping variable really a string variable that uses numeric codes? If so, then you would need to quote the values:

t-test groups=groupvar ('1', '2')
  /variables=testvar.

But if it's really numeric with defined value labels (e.g., 1="No", 2="Yes"), the values should not be quoted on the groups subcommand, and make sure you're using the underlying values not the value labels:

t-test groups=groupvar (1, 2)
  /variables=testvar.

Finally, if the grouping variable is a string with values such as 'n' and 'y', then case matters. For example:

t-test groups=groupvar ('Y', 'N')
  /variables=testvar.

will probably give you the error to which you refer if the actual string data values are all entered as lower case "y" and "n".

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Richard Ristow
Sent: Friday, June 29, 2007 11:01 AM
To: [hidden email]
Subject: Re: Confusion with Independent Samples T Test

At 10:31 AM 6/29/2007, holly costar wrote:

>I'm getting warnings when trying to execute an independent samples t
>test. The warnings say that one of the groups is empty and the test
>can't be executed. I've defined the groups (using a string variable as
>grouping, y/n coded 1/2 for whether or not participants have health
>insurance) and there are no gaps in the data. Any idea what I'm doing
>wrong?

No idea offhand, but here are some quick diagnostic tools:

- Run FREQUENCIES on your grouping variable. If there are no counts in
one category, that's the problem. Then, it may be,
. You defined the grouping variable using SPSS transformations, and
there's a bug in your logic - so, check the logic
. You don't have the population you thought you had (sorry, but this
does happen). To check that, run descriptive statistics that will
characterize your data relative to the population you expected
. If you do see members in both groups, perhaps you specified the
categories wrong on the /GROUPS subcommand

Forgive me if these sound too simple-minded. The fact is, however
experienced we are, a lot of our mistakes are something very simple.

- Run MEANS, for your independent variable by your grouping variable.
Of course, if T-TEST finds a category empty, so will MEANS. But
. If FREQUENCIES shows members in a group and MEANS doesn't, then all
values of the dependent are missing in that group. That can come from
many sources, including logic errors and oddities in how the data was
gathered in the first place. If there are few valid values for any
group, your dependent variable is sparse, arising either from a mistake
in data-gathering or computation; or data that really *is* sparse,
which usually takes a little rethinking of the study.

Wild shots, of course. But these are what I'd do first, given your
situation.

-Best of luck,
  Richard Ristow