Testing Opinions between Schools

I have about 300 cases whereby respondents indicated Familiarity with each
of 10 schools (5-point scale) then they are asked a series of opinion
questions for each of the schools (5-point or 7-point scales depending on
the question)

I would like to compare 3 of the schools based on top 2-box scores.

I created a "school score" depending on whether or not the respondent scored
the school in the top 2-box for the Familiarity question and this works
nicely BUT a respondent could be scored into all of the 3 schools since they
were at least familiar with multiple schools: Not mutually exclusive scoring
so not a differentiator to do tests of comparison (at least it appears so to
me)

My question: Does this school scoring sound viable or would another
comparison method be more appropriate

Hope this is clear?

Tks
Will

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

Well, and I thought it was 'reasonably' clear

 

Thanks Rich, trying again

 

The bottom line is that this is a basic comparison test of mean differences between schools A v B v C (see a Table with schools as column headings and the Likert question means as rows), however, in this case the column headings aren't mutually exclusive like age groups or income. School would be the grouping variable in a test like the independent t-test if mutually exclusive.

 

There isn't a distinct code for each school it had to be developed using another question's result (e.g., the Familiar question). So, in this case, a respondent could be familiar with multiple schools which voids the mutually exclusive need for the grouping variable.

 

My questions are: Am I correct in that the column headings be mutually exclusive? If so, what other test options do I have to compare the Likert questions between schools given the way I had to create the school segment.

 

Tks

---

From: Rich Ulrich [mailto:[hidden email]]
Sent: Saturday, October 22, 2011 12:38 AM
To: [hidden email]; SPSS list
Subject: RE: Testing Opinions between Schools

See comments as inserted -

> Date: Fri, 21 Oct 2011 10:07:39 -0400

> From: [hidden email]
> Subject: Testing Opinions between Schools
> To: [hidden email]
>
> I have about 300 cases whereby respondents indicated Familiarity with each
> of 10 schools (5-point scale) then they are asked a series of opinion
> questions for each of the schools (5-point or 7-point scales depending on
> the question)
>
> I would like to compare 3 of the schools based on top 2-box scores.

The sense here is difficult to figure.

If you said you wanted to compare "3 of the schools based on Long Island",
I would know that this was a selection of schools.  If you left out the
word "based", I would know you were comparing those scores, even though
I don't know what a "top 2-box score" is.  

>
> I created a "school score" depending on whether or not the respondent scored
> the school in the top 2-box for the Familiarity question and this works
> nicely

 - please - works nicely to do what?  works nicely in what way?
Is this 10 scores, one for each school?
Does it serve to find 3 schools with a whole lot of ratings?

>         BUT a respondent could be scored into all of the 3 schools since they
> were at least familiar with multiple schools: Not mutually exclusive scoring
> so not a differentiator to do tests of comparison (at least it appears so to
> me)
>
> My question: Does this school scoring sound viable or would another
> comparison method be more appropriate

 (a) I can't tell what "this school scoring" is, 
 (b) what it is that you are referring to as a comparison method, and
 (c) "appropriate" for what?

>
> Hope this is clear?

Nope.  And I'm usually pretty good at getting the point.

--
Rich Ulrich

[Rich Ulrich]

Here is an example of something that is reasonably clear, and
it does describe what you have, as I read between the lines of
both messages.  If this does not describe what you have, you
can try again.

There are 300 "cases" with ratings of 10 schools by each rater. 
This says that you have either 300 respondents rating 10, or 30
respondents rating 10 ... but that is relatively unimportant for now.

The raw data looks something like this.

Rater School Familiarity score1 to scoreNN
001    AAA        3    < scores to analyze >
001    BBB         4    ..
 - 10 lines for rater 001; similar for raters 002-030 or 002 to 300.
 - down to
030    JJJ           5   ..


A straightforward analysis that does not take into account
familiarity would be a repeated-measures ANOVA on the
interesting composite scores created from score1 to scoreNN;
or - even - on some of the items.  That analysis could use
all 10 lines by a rater as one record; or it could be set up as
a 2-way design, Rater x School. 

You propose to use "familiarity" to reduce the file size by excuding
every report where that is not high enough.  That leaves an unbalanced-
blocks design, so probably want to read the 2-5 records for each
rater as separate records.  Then you do the same 2-way design.
You would want to look at and report the  means,
rather than the raw means.

Use *all*  the data with high familiarity.  If you are only interested
in contrasts concerning three of the schools, you can specify
those contrasts for particular testing.


If you analyze the omitted records in the same way, you have a
parallel analysis that reports on something like "popular opinion"
of how the schools rate.

--
Rich Ulrich

---

From: [hidden email]
To: [hidden email]
CC: [hidden email]
Subject: RE: Testing Opinions between Schools
Date: Sat, 22 Oct 2011 08:44:37 -0400

Well, and I thought it was 'reasonably' clear

 

Thanks Rich, trying again

 

The bottom line is that this is a basic comparison test of mean differences between schools A v B v C (see a Table with schools as column headings and the Likert question means as rows), however, in this case the column headings aren't mutually exclusive like age groups or income. School would be the grouping variable in a test like the independent t-test if mutually exclusive.

 

There isn't a distinct code for each school it had to be developed using another question's result (e.g., the Familiar question). So, in this case, a respondent could be familiar with multiple schools which voids the mutually exclusive need for the grouping variable.

 

My questions are: Am I correct in that the column headings be mutually exclusive? If so, what other test options do I have to compare the Likert questions between schools given the way I had to create the school segment.

 

Tks
[snip, previous]

[Beckstead, Jason]

I will be away at conference Oct 22 through Oct 26.

[Allum, Jeff]

I will be out of the office on Monday, October 24. I will reply to your e-mail after I return on Tuesday, October 25.

Jeff
____________________________
Jeff Allum
Research Associate
Council of Graduate Schools
One Dupont Circle, NW, Suite 230
Washington, DC 20036-1173
[phone] (direct)
[phone] (main)
[phone] (fax)
[hidden email]<mailto:[hidden email]>
www.cgsnet.org<http://www.cgsnet.org/>

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

[zstatman]

Thanks Rich, let me think about this since the schools are not uniquely identified as a variable as you illustrate

 

W

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Rich Ulrich
Sent: Saturday, October 22, 2011 7:32 PM
To: [hidden email]
Subject: Re: Testing Opinions between Schools

Here is an example of something that is reasonably clear, and
it does describe what you have, as I read between the lines of
both messages.  If this does not describe what you have, you
can try again.

There are 300 "cases" with ratings of 10 schools by each rater. 
This says that you have either 300 respondents rating 10, or 30
respondents rating 10 ... but that is relatively unimportant for now.

The raw data looks something like this.

Rater School Familiarity score1 to scoreNN
001    AAA        3    < scores to analyze >
001    BBB         4    ..
 - 10 lines for rater 001; similar for raters 002-030 or 002 to 300.
 - down to
030    JJJ           5   ..


A straightforward analysis that does not take into account
familiarity would be a repeated-measures ANOVA on the
interesting composite scores created from score1 to scoreNN;
or - even - on some of the items.  That analysis could use
all 10 lines by a rater as one record; or it could be set up as
a 2-way design, Rater x School. 

You propose to use "familiarity" to reduce the file size by excuding
every report where that is not high enough.  That leaves an unbalanced-
blocks design, so probably want to read the 2-5 records for each
rater as separate records.  Then you do the same 2-way design.
You would want to look at and report the  means,
rather than the raw means.

Use *all*  the data with high familiarity.  If you are only interested
in contrasts concerning three of the schools, you can specify
those contrasts for particular testing.


If you analyze the omitted records in the same way, you have a
parallel analysis that reports on something like "popular opinion"
of how the schools rate.

--
Rich Ulrich

---

From: [hidden email]
To: [hidden email]
CC: [hidden email]
Subject: RE: Testing Opinions between Schools
Date: Sat, 22 Oct 2011 08:44:37 -0400

Well, and I thought it was 'reasonably' clear

 

Thanks Rich, trying again

 

The bottom line is that this is a basic comparison test of mean differences between schools A v B v C (see a Table with schools as column headings and the Likert question means as rows), however, in this case the column headings aren't mutually exclusive like age groups or income. School would be the grouping variable in a test like the independent t-test if mutually exclusive.

 

There isn't a distinct code for each school it had to be developed using another question's result (e.g., the Familiar question). So, in this case, a respondent could be familiar with multiple schools which voids the mutually exclusive need for the grouping variable.

 

My questions are: Am I correct in that the column headings be mutually exclusive? If so, what other test options do I have to compare the Likert questions between schools given the way I had to create the school segment.

 

Tks
[snip, previous]