Randomly selecting participants until sample size equals an integer, then repeating this 1000 times prior to analysis

[Philipp802]

Hey everyone,

I have the following question for the following data set. I have 5
conditions with 1 variable of interest (scale 0-20). All conditions have
both men and women, but there are more women than men in all conditions. I
do find a statistically significant effect between genders in some of these
conditions, but I would like to investigate if this effect would also hold
true if the genders were evenly distributed.

So I aim to randomly select female participants in each condition up until
their n is equal to the male n (without replacement). Then I'd like to do
that 1000 times and then run the analysis again.

Is there any way to do that in SPSS?

Thank you very much for your help!

Best,
Philipp



--
Sent from: http://spssx-discussion.1045642.n5.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

Why not weighting your cases so that both groups are equally sized within your treatments?

Mario Giesel

Munich, Germany

Am Donnerstag, 12. November 2020, 11:36:40 MEZ hat Philipp802 <[hidden email]> Folgendes geschrieben:

Hey everyone,

I have the following question for the following data set. I have 5

conditions with 1 variable of interest (scale 0-20). All conditions have

both men and women, but there are more women than men in all conditions. I

do find a statistically significant effect between genders in some of these

conditions, but I would like to investigate if this effect would also hold

true if the genders were evenly distributed.

So I aim to randomly select female participants in each condition up until

their n is equal to the male n (without replacement). Then I'd like to do

that 1000 times and then run the analysis again.

Is there any way to do that in SPSS?

Thank you very much for your help!

Best,

Philipp

--

Sent from: http://spssx-discussion.1045642.n5.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

[Jon Peck]

You are, of course, sacrificing power by doing this.  You might consider using the Welch variation if robustness is your concern as well a bootstrapping.

However, if you  want to pursue your original thought, you can see how to draw random samples of an  exact size by doing Data > Select Cases > Random Sample > Exact size.

You would sort the sample by gender and then specify the size to sample from based on the male n.

To automate that over many samples, you would need to use a little Python code combined with this process.

On Thu, Nov 12, 2020 at 3:36 AM Philipp802 <[hidden email]> wrote:

Hey everyone,

I have the following question for the following data set. I have 5
conditions with 1 variable of interest (scale 0-20). All conditions have
both men and women, but there are more women than men in all conditions. I
do find a statistically significant effect between genders in some of these
conditions, but I would like to investigate if this effect would also hold
true if the genders were evenly distributed.

So I aim to randomly select female participants in each condition up until
their n is equal to the male n (without replacement). Then I'd like to do
that 1000 times and then run the analysis again.

Is there any way to do that in SPSS?

Thank you very much for your help!

Best,
Philipp



--
Sent from: http://spssx-discussion.1045642.n5.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

[Rich Ulrich]

Your statement is hard to figure out,

  "I would like to investigate if this effect would also hold
   true if the genders were evenly distributed."

You say that Ns by gender are unequal; you do not say that they are

unbalanced between the 5 groups.  Are they unbalanced (not equal,

and not proportionate)?

Or, maybe, are you doing t-tests with varying sample sizes, and

wondering if the different outcomes owe to different Ns?

Throwing away cases is hard to justify, so, don't do it.

Doing it multiple times, sort of like a bootstrap, is also hard to

justify.  I don't think you will find editor or audience that will like

your ad-hoc idea for that, either.

For formal hypothesis testing, your 5 groups by 2 sexes seems suited

to a two-way ANOVA which tests the two main effects.  The default

assumption would be that any main effect by sex exists in every group.

They are tested together because we desire compact logic and the good

power it provides to testing.

Testing for the separate groups might be justified by finding a significant

test on the 4 d.f.  interaction term of the ANOVA.  Or -  Do you have a

reason to expect effects in certain groups and not in others?

If you are exploring and not maintaining a formal control for hypothesis

testing, just do the t-tests. 

--

Rich Ulrich

---

From: SPSSX(r) Discussion <[hidden email]> on behalf of Philipp802 <[hidden email]>
Sent: Thursday, November 12, 2020 5:36 AM
To: [hidden email] <[hidden email]>
Subject: Randomly selecting participants until sample size equals an integer, then repeating this 1000 times prior to analysis

 

Hey everyone,

I have the following question for the following data set. I have 5
conditions with 1 variable of interest (scale 0-20). All conditions have
both men and women, but there are more women than men in all conditions. I
do find a statistically significant effect between genders in some of these
conditions, but I would like to investigate if this effect would also hold
true if the genders were evenly distributed.

So I aim to randomly select female participants in each condition up until
their n is equal to the male n (without replacement). Then I'd like to do
that 1000 times and then run the analysis again.

Is there any way to do that in SPSS?

Thank you very much for your help!

Best,
Philipp



--
Sent from: http://spssx-discussion.1045642.n5.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