SPSSX Discussion

Question on effect sizes

Classic

List

Threaded

4 messages Options

Tanja Gabriele Baudson

Question on effect sizes

Hi all,

what effect size would be best suited for (M)ANOVAs? The SPSS default
is eta^2; is there anything better (comparable to Cohen's d) for
multivariate designs?

Thanks in advance, and have a nice weekend!
Tanya

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

Robert L

Stratified sample?

I have got this problem which wouldn't be much of a problem with a better understanding of SPSS. A simplified description:

*The basic file contains 1000 individuals of different ages, 35, 50 and 65 years old living in two regions. A crosstab would show the following:

35 50 65
A 200 100 50
B 300 200 150

*The goal is to sample 50 from each region in a way that corresponds to the age distribution in a given region. In region A that would mean a sample of 200/350*50=28.6 or 29 individuals of age 35, 14 indiviudals of age 50 and 7 of age 65.

*This procedure should be repeated for each region.

As this will be a fairly common task with more ages and more regions, I would prefer not to do the sampling manually but rather find a way to set up a nifty syntax file. This must be simple, and I suspect the best way would be a combination of DO IF, DO REPEAT and SAMPLE commands. It could possibly presumably also be achieved with the Complex Samples module which is not included in our installation. And the calculations needed are not complex enough to motivate the addition of this module. It is not so much a question of knowledge about sampling procedures as basic knowledge of SPSS basics which I really should know... Suggestions would be most welcome.

Robert
******************
Robert Lundqvist
Norrbotten regional council
Lulea, Sweden

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

Robert Lundqvist

Art Kendall

Re: Stratified sample?

Something like the untested syntax below should do what you say.
However, be sure that you have sufficient cases for the effects you want to test for. Also finite pop corrections are questionable if you want to generalize outside the given time, etc.
ANOVA type comparisons are usually better with equal Ns. Weighted selecting optimizes the precision of the grand total (mean).
The sample sizes from a priori estimates is for achieved sample size. That is completed cases. Case nonresponse can be very large.
Often the cost of gathering and processing data per case sampled is sufficiently small relative to the total cost of the project that erring on the side of "too large" samples often results in little change in the total cost of the project.
Some costs that are relatively fixed are design consultation, frame acquisition, instrument design, pre-testing, instrument revision, computer time, analyst time, software costs, lines of syntax, data entry specifications, report writing, and report reviewing
Some costs that increase per case sampled are postage, printing, interviewer time, interviewer travel, data entry.
A costs that increases slightly per case sampled is data cleaning time. More problems will be found in the data, but most of the cost is developing the syntax to find the problems.

Do if Region eq 'A'. + Do if agegroup eq 35. + Select 29 from 200 + Else if agegroup eq 50. + Select nn from 100. + Else if agegroup eq 65. + Select nn from 50. + Else. + Print /' oops in region A'. + end if. Do if Region eq 'B'. + Do if agegroup eq 35. + Select nn from 300 + Else if agegroup eq 50. + Select nn from 200. + Else if agegroup eq 65. + Select nn from 150. + Else. + Print /' oops in region B'. + end if.

Art Kendall
Social Research Consultants

On 12/3/2010 9:19 AM, Robert Lundqvist wrote:

I have got this problem which wouldn't be much of a problem with a better understanding of SPSS.  A simplified description:

*The basic file contains 1000 individuals of different ages, 35, 50 and 65 years old living in two regions. A crosstab would show the following:

        35      50      65
A       200     100     50
B       300     200     150

*The goal is to sample 50 from each region in a way that corresponds to the age distribution in a given region. In region A that would mean a sample of 200/350*50=28.6 or 29 individuals of age 35, 14 indiviudals of age 50 and 7 of age 65.

*This procedure should be repeated for each region.

As this will be a fairly common task with more ages and more regions, I would prefer not to do the sampling manually but rather find a way to set up a nifty syntax file. This must be simple, and I suspect the best way would be a combination of DO IF, DO REPEAT and SAMPLE commands. It could possibly presumably also be achieved with the Complex Samples module which is not included in our installation. And the calculations needed are not complex enough to motivate the addition of this module. It is not so much a question of knowledge about sampling procedures as basic knowledge of SPSS basics which I really should know...  Suggestions would be most welcome.

Robert
******************
Robert Lundqvist
Norrbotten regional council
Lulea, Sweden

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

Art Kendall
Social Research Consultants

ViAnn Beadle

Re: Stratified sample?

What the OP is looking for is raking or sometimes called sample balancing. IIRC, there is a python module on DevCentral which does this.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Art Kendall
Sent: Friday, December 03, 2010 8:13 AM
To: [hidden email]
Subject: Re: Stratified sample?

Something like the untested syntax below should do what you say.
However, be sure that you have sufficient cases for the effects you want to test for. Also finite pop corrections are questionable if you want to generalize outside the given time, etc.
ANOVA type comparisons are usually better with equal Ns. Weighted selecting optimizes the precision of the grand total (mean).
The sample sizes from a priori estimates is for achieved sample size. That is completed cases. Case nonresponse can be very large.
Often the cost of gathering and processing data per case sampled is sufficiently small relative to the total cost of the project that erring on the side of "too large" samples often results in little change in the total cost of the project.
Some costs that are relatively fixed are design consultation, frame acquisition, instrument design, pre-testing, instrument revision, computer time, analyst time, software costs, lines of syntax, data entry specifications, report writing, and report reviewing
Some costs that increase per case sampled are postage, printing, interviewer time, interviewer travel, data entry.
A costs that increases slightly per case sampled is data cleaning time. More problems will be found in the data, but most of the cost is developing the syntax to find the problems.

Do if Region eq 'A'.
+    Do if agegroup eq 35.
+    Select 29 from 200
+    Else if agegroup eq 50.
+    Select nn from 100.
+    Else if agegroup eq 65.
+    Select nn from 50.
+    Else.
+    Print /' oops in region A'.
+    end if.
Do if Region eq 'B'.
+    Do if agegroup eq 35.
+    Select nn from 300
+    Else if agegroup eq 50.
+    Select nn from 200.
+    Else if agegroup eq 65.
+    Select nn from 150.
+    Else.
+    Print /' oops in region B'.
+    end if.

Art Kendall
Social Research Consultants

On 12/3/2010 9:19 AM, Robert Lundqvist wrote:

I have got this problem which wouldn't be much of a problem with a better understanding of SPSS.  A simplified description:

*The basic file contains 1000 individuals of different ages, 35, 50 and 65 years old living in two regions. A crosstab would show the following:

        35      50      65

A       200     100     50

B       300     200     150

*The goal is to sample 50 from each region in a way that corresponds to the age distribution in a given region. In region A that would mean a sample of 200/350*50=28.6 or 29 individuals of age 35, 14 indiviudals of age 50 and 7 of age 65.

*This procedure should be repeated for each region.

As this will be a fairly common task with more ages and more regions, I would prefer not to do the sampling manually but rather find a way to set up a nifty syntax file. This must be simple, and I suspect the best way would be a combination of DO IF, DO REPEAT and SAMPLE commands. It could possibly presumably also be achieved with the Complex Samples module which is not included in our installation. And the calculations needed are not complex enough to motivate the addition of this module. It is not so much a question of knowledge about sampling procedures as basic knowledge of SPSS basics which I really should know...  Suggestions would be most welcome.

Robert

******************

Robert Lundqvist

Norrbotten regional council

Lulea, Sweden

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

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