TURF analysis question

[Dave Taber]

i'm using the TURF analysis utility that John Peck programmed and am wondering if anyone has come up with a way to add this feature

easiest to illustrate with an example

supposed we showed respondents a list of 10 new TV shows for a new network and asked them whether are not that would watch each show -- currently the TURF analysis tells us what combination of shows would provide the highest % of people watching at least one show (reach) -- it can do this for any set number of shows from 1 to 10

but let's assume the the new network believes that people need to watch at least 2 shows on it's network before that consider a viewer to be a loyal viewer so the analysis they want to run is to determine which combination of n shows brings in the highest percentage of people who would watch at least 2 (or at least n) shows

any thoughts on how to accomplish this within SPSS?

thanks
Dave

[zstatman]

Jon monitors this listserve so keep an eye open

 

WMB
Statistical Services

============
mailto: [hidden email]
http:\\home.earthlink.net\~info.statman
============

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Dave Taber
Sent: Friday, October 09, 2009 2:43 PM
To: [hidden email]
Subject: TURF analysis question

i'm using the TURF analysis utility that John Peck programmed and am wondering if anyone has come up with a way to add this feature

easiest to illustrate with an example

supposed we showed respondents a list of 10 new TV shows for a new network and asked them whether are not that would watch each show -- currently the TURF analysis tells us what combination of shows would provide the highest % of people watching at least one show (reach) -- it can do this for any set number of shows from 1 to 10

but let's assume the the new network believes that people need to watch at least 2 shows on it's network before that consider a viewer to be a loyal viewer so the analysis they want to run is to determine which combination of n shows brings in the highest percentage of people who would watch at least 2 (or at least n) shows

any thoughts on how to accomplish this within SPSS?

thanks
Dave

[Peck, Jon]

Can you solve this by just selecting out only cases with two or more yes responses and then running TURF?

The COUNT command can add those responses up for you.

 

HTH,

Jon

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Dave Taber
Sent: Friday, October 09, 2009 12:43 PM
To: [hidden email]
Subject: [SPSSX-L] TURF analysis question

 

i'm using the TURF analysis utility that John Peck programmed and am wondering if anyone has come up with a way to add this feature

easiest to illustrate with an example

supposed we showed respondents a list of 10 new TV shows for a new network and asked them whether are not that would watch each show -- currently the TURF analysis tells us what combination of shows would provide the highest % of people watching at least one show (reach) -- it can do this for any set number of shows from 1 to 10

but let's assume the the new network believes that people need to watch at least 2 shows on it's network before that consider a viewer to be a loyal viewer so the analysis they want to run is to determine which combination of n shows brings in the highest percentage of people who would watch at least 2 (or at least n) shows

any thoughts on how to accomplish this within SPSS?

thanks
Dave

[Humphrey Paulie]

Dear colleagues,

In a study a researcher investigates the effect of computer anxiety on computer based test performance. He has developed a test and delivered it in two modes: paper and pencil (PP) and computer-based (CBT). The two tests are exactly the same only the delivery mode is different. A group of 40 subjects has taken both tests. He has also divided the subjects into 3 anxiety levels: low, mid and high.

He has run a two-way ANOVA to investigate the effects of computer anxiety on test performance and the interaction between delivery mode and anxiety.

In fact there are two categorical independent variables: delivery mode (PP and CBT) and anxiety levels (mid, low, high) and one continuous dependent variable: performance on the test (not sure whether it is scores on the PP or CBT). Is this design ok? Is it fine when the subjects of the two categories of PP and CBT are the same? Should the researcher consider scores on the PP as the dependent variable or scores on the CBT?

Is there an alternative design?

Id be grateful for comments.

Humphrey

 

 

 

 

[Granaas, Michael]

This looks suspiciously like a homework problem.  I'd like I bit more info before commenting.

 

Michael

****************************************************
Michael Granaas             [hidden email]
Assoc. Prof.                Phone: [phone]
Dept. of Psychology         FAX:  [phone]
University of South Dakota
414 E. Clark St.
Vermillion, SD 57069
*****************************************************

---

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Humphrey Paulie [[hidden email]]
Sent: Friday, October 09, 2009 4:09 PM
To: [hidden email]
Subject: Two-way ANOVA

Dear colleagues,

In a study a researcher investigates the effect of computer anxiety on computer based test performance. He has developed a test and delivered it in two modes: paper and pencil (PP) and computer-based (CBT). The two tests are exactly the same only the delivery mode is different. A group of 40 subjects has taken both tests. He has also divided the subjects into 3 anxiety levels: low, mid and high.

He has run a two-way ANOVA to investigate the effects of computer anxiety on test performance and the interaction between delivery mode and anxiety.

In fact there are two categorical independent variables: delivery mode (PP and CBT) and anxiety levels (mid, low, high) and one continuous dependent variable: performance on the test (not sure whether it is scores on the PP or CBT). Is this design ok? Is it fine when the subjects of the two categories of PP and CBT are the same? Should the researcher consider scores on the PP as the dependent variable or scores on the CBT?

Is there an alternative design?

Id be grateful for comments.

Humphrey

 

 

 

 

[Dave Taber]

you can get the TURF analysis utility on the spss website:

http://www.spss.com/devcentral/index.cfm?pg=plugins

also more info on it from an article written by Jon Peck here: http://insideout.spss.com/2009/03/03/python-and-productivity/

On Sat, Oct 10, 2009 at 12:44 PM, Pushpender Nath <[hidden email]> wrote:

Hi Dave

Could you please share this utility for me. I need it in my work.

On Sat, Oct 10, 2009 at 12:13 AM, Dave Taber <[hidden email]> wrote:

i'm using the TURF analysis utility that John Peck programmed and am wondering if anyone has come up with a way to add this feature

easiest to illustrate with an example

supposed we showed respondents a list of 10 new TV shows for a new network and asked them whether are not that would watch each show -- currently the TURF analysis tells us what combination of shows would provide the highest % of people watching at least one show (reach) -- it can do this for any set number of shows from 1 to 10

but let's assume the the new network believes that people need to watch at least 2 shows on it's network before that consider a viewer to be a loyal viewer so the analysis they want to run is to determine which combination of n shows brings in the highest percentage of people who would watch at least 2 (or at least n) shows

any thoughts on how to accomplish this within SPSS?

thanks
Dave

--


Regards

Pushpender Nath

[Jarrod Teo-2]

Hi,

I need clarification for something basic.

We know that 2 independent samples T-test has the following assumptions

1. The scale follows normal distribution.
2. The 2 categories in the categorical variable is mutually exclusive.
3. The homogenity of variance assumption has to be met.

My question is as follows

1. Assuming that the normality is met, what happen if the homogenity of variance assumption failed? Do we go to a non-parametric test say Mann Whitney or do we just use the variance not assumed T-test Statistics in the T-test table?
2. I had a professor once commented that he will use the variance not assumed T-test Statistics and that Statistics is the calculation when the variance is not pooled but for separated categories. He even mentioned that going from Parametric to Non-parametrice approach is not correct. Is he right by commenting these?

Thanks in advance.

Regards
Dorraj

---

Windows Live: Keep your friends up to date with what you do online.

[Marta Garcia-Granero]

Hi Dorraj

DorraJ Oet wrote:

>
> I need clarification for something basic.
>
> We know that 2 independent samples T-test has the following assumptions
>
>    1. The scale follows normal distribution.
>    2. The 2 categories in the categorical variable is mutually exclusive.
>    3. The homogenity of variance assumption has to be met.
>
>
> My question is as follows
>
>    1. Assuming that the normality is met, what happen if the
>       homogenity of variance assumption failed? Do we go to a
>       non-parametric test say Mann Whitney or do we just use the
>       variance not assumed T-test Statistics in the T-test table?
>

Although not a lot of people knows that, Mann-Whitney's test is very
sensitive to differences in spread (heterogeneity of variances).
Therefore, it is as bad as a T-test when homogeneity of variances is not
met. You must use the "variances not assumed" (Welch test is the name,
BTW) T-test.
>
>    1. I had a professor once commented that he will use the variance
>       not assumed T-test Statistics and that Statistics is the
>       calculation when the variance is not pooled but for separated
>       categories. He even mentioned that going from Parametric to
>       Non-parametrice approach is not correct. Is he right by
>       commenting these?
>

Yes, he is indeed (see above). "Distribution-free" is a misnomer.
Non-parametric tests can have some conditions concerning the
distribution of the data. For instance, signed ranks Wilcoxon test needs
symmetrical distributions, Mann-Whitney's U and Kruskal-Wallis tests are
sensitive to differences in shape and or spread of the groups being
compared...

HTH,
Prof. Marta García-Granero

--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

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

[Bruce Weaver]

DorraJ wrote

Hi,

I need clarification for something basic.

We know that 2 independent samples T-test has the following assumptions

The scale follows normal
     distribution.The 2 categories in the
     categorical variable is mutually exclusive.The homogenity of variance
     assumption has to be met.

My question is as follows

Assuming that the normality
     is met, what happen if the homogenity of variance assumption failed? Do we
     go to a non-parametric test say Mann Whitney or do we just use the variance not assumed T-test Statistics in the T-test table?I had a professor once commented that he will use the variance not assumed T-test Statistics and that Statistics is the calculation when the variance is not pooled but for separated categories. He even mentioned that going from Parametric to Non-parametrice approach is not correct. Is he right by commenting these?
Thanks in advance.

Regards
Dorraj

You omitted the most important assumption for the unpaired t-test, i.e., independence of observations.  You did  say two mutually exclusive categories, but that is not the same thing as independence of observations (both between and within categories).  

As for normality & homogeneity of variance, these assumptions are never truly met, at least when one has real data (as opposed to simulated data).  I wish that textbooks were clearer on this.  What they should say, IMO, is something like the following:

If one could sample from two populations that have exactly equal means and variances, and if all scores were mutually independent, then the t-test (as computed with the usual formula) would have a sampling distribution that is given exactly by the t-distribution with df = n1 + n2 - 2.  But since the normality and homogeneity of variance assumptions are never met with real data, the t-test is an approximate test, not an exact test.  And therefore, the important question is whether the conditions are such that the approximation is a useful one.  (Remember what George Box said about models--i.e., they're all wrong, but some of them are useful.)

In this context, I'll add something to the good response you got from Marta.  How worried you need to be about heterogeneity of variance depends on the similarity of the two sample sizes.  When sample sizes are equal, the t-test (and ANOVA) are quite robust to heterogeneity of variance.  One rule of thumb says that with equal sample sizes, the ratio of largest to smallest variances can be as large as 4 or 5 (see Dave Howell's book Statistical Methods for Psychology, for example).  This is why folks who do designed experiments usually strive to keep the group sizes as balanced as possible.

[Marta Garcia-Granero]

Bruce Weaver wrote:

> ...
>
>
> In this context, I'll add something to the good response you got from Marta.
> How worried you need to be about heterogeneity of variance depends on the
> similarity of the two sample sizes.  When sample sizes are equal, the t-test
> (and ANOVA) are quite robust to heterogeneity of variance.  One rule of
> thumb says that with equal sample sizes, the ratio of largest to smallest
> variances can be as large as 4 or 5 (see Dave Howell's book Statistical
> Methods for Psychology, for example).  This is why folks who do designed
> experiments usually strive to keep the group sizes as balanced as possible.
>

The rule of thumb used in the book "Statistics at Square One" (p. 69-70)
is very similar to the one you mention: the ratio of the larger to the
smaller standard deviation should not be greater than 2. This is the
same that stating that the ratio of variances should not be greater than 4.


Best regards,
Marta GG

--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

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