Fw: Excluded variables from regression

[SPSS Support-2]

In my earlier response, I had inferred from Caroline's description that her Coefficients table was OK. Looking at the description more closely in the light of Bruce's response, I would expect all of the product terms to appear in the Model 1 subtable of the Excluded variables table and one of the six product terms to appear in the Model 2 subtable. With six groups, the six product terms would sum to the continuous predictor and one of them would be redundant. As Bruce stated, the 5 product terms entered in Block 2 should correspond to the same 5 groups as the dummy variables in Block 1.

David Matheson
Statistical Support
SPSSS, an IBM company    

From:	Bruce Weaver <[hidden email]>
To:	[hidden email]
Date:	01/07/2010 12:59 PM
Subject:	Re: Excluded variables from regression
Sent by:	"SPSSX(r) Discussion" <[hidden email]>

---

CarolineUK wrote:
>
> This question was asked last year...
>
> Does anyone know why SPSS excludes certain variables from a regression?
> Also, does anyone know how to prevent this from happening?
>
> There were various offers of help/suggestions:
> that the variables excluded were not numeric
> That stepwise had been used instead of 'enter'
> That there was multicollinearity
>
> I've checked my variables and they contain numeric values, I'm using
> 'enter'
> and the tolerance levels are at least .4 or above.
>
> The background: I'm using hierarchical multiple regression to check for
> interaction between a qualitative continuous predictor variable and group
> membership.  The continuous predictor variable and dummy variables to
> denote
> group (coded 0 for not in group, 1 for in group) go in block one, along
> with
> some covariate continuous predictor variables.  Then each of the six dummy
> variables is linked with one of the continuous predictor variables to
> create
> six product terms and these go in block two.
>
> (For clarity, I'm leaving one group out of the first block and this will
> be
> represented by the constant, but putting all six product terms in block
> two)
>
>
--- snip ---

If I follow, there are 6 groups that are coded with 5 indicator variables in
block 1.  If that is correct, then the interaction of that GROUP variable
with the continuous variable needs to be represented by 5 product terms, not
6.  The 5 product terms are the same 5 indicator variables you have in block
1 multiplied by the continuous variable.



-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/
"When all else fails, RTFM."

NOTE:  My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.
--
View this message in context: http://old.nabble.com/Excluded-variables-from-regression-tp14957806p27065134.html
Sent from the SPSSX Discussion mailing list archive at 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

[Poling, Taylor Leigh]

 

Hi List,

 

Can anyone suggest some syntax for flagging cases where responses are all the same across a set of many variables(i.e., representing survey items) without having to write a giant, complex IF statement?

 

Thanks,

Taylor

[Zdaniuk, Bozena-3]

You can use COUNT command. For example, if you want to know who said yes=2 to variables A thru D, you can do:

 

COUNT COUNTYES=VARA TO VARD (2).

EXE.

 

And then the cases with COUNTYES=4 are those that have answer yes=2 to each of the four variables.

Bozena

 

Bozena Zdaniuk, Ph.D.

Human Early Learning Partnership (HELP)
College for Interdisciplinary Studies
UBC - UVic - SFU - UNBC - TRU - UBCO
440 - 2206 East Mall, Vancouver BC  V6T 1Z3
Tel: [phone] / Fax: [phone]
www.earlylearning.ubc.ca

 

From: Poling, Taylor L [mailto:[hidden email]]
Sent: Thursday, January 07, 2010 1:07 PM
To: [hidden email]
Subject: sytax for no variance in survey item responding

 

 

Hi List,

 

Can anyone suggest some syntax for flagging cases where responses are all the same across a set of many variables(i.e., representing survey items) without having to write a giant, complex IF statement?

 

Thanks,

Taylor

[Maguin, Eugene]

Taylor,

Look at the SD.n function as in

Compute y=sd.1(x1 to x367).

If y computes to 0.00, all your variables for some case have the same value.



Gene Maguin


>>Can anyone suggest some syntax for flagging cases where responses are all
the same across a set of many variables(i.e., representing survey items)
without having to write a giant, complex IF statement?

Thanks,
Taylor

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

[Marks, Jim]

Taylor:

 

The easiest way would be use the count function.

 

not tested.

 

**find out how many answers in var1 to var20 have a value of 7 thru 1.

COUNT sevens = var1 TO var20 (7)

   /sixes = var1 TO var20 (6)

   /fives= var1 TO var20 (5)

   /fours = var1 TO var20 (4)

   /threes = var1 TO var20 (3)

   /twos = var1 TO var20 (2)

   /ones = var1 TO var20 (1)

.

 

** mark a case as suspicious if 15 or more answers are the same.

COMPUTE suspicious = ANY(15, sevens TO ones).

 

EXECUTE.

 

Note that this syntax uses the TO convention to specify all the variables between var1 and var20. You might need to list out the variables to be checked if they are not contiguous in the dataset.

 

Jim Marks

Director, Market Research

x1616

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Poling, Taylor L
Sent: Thursday, January 07, 2010 3:07 PM
To: [hidden email]
Subject: sytax for no variance in survey item responding

 

 

Hi List,

 

Can anyone suggest some syntax for flagging cases where responses are all the same across a set of many variables(i.e., representing survey items) without having to write a giant, complex IF statement?

 

Thanks,

Taylor

[John F Hall]

35 years ago I faced a similar problem using an 11 point 0 (completely dissatisfied) to 10 (completely satisfied) scale on a hundred or so domains and sub-domains in the Quality of Life in Britain surveys SSRC Survey Unit Quality of Life Surveys I did with the late Dr Mark Abrams (SSRC Survey Unit) and in collaboration with the late Prof Angus Campbell (ISR, Ann Arbor).  We had done a pilot survey using 0-10 and another using a 1-7 scale (as per the USA survey) in which responses were highly (negatively) skewed and so we switched back to 0-10 to get ratings to spread out more, but this created a different problem.  The fullest account of these surveys is Subjective measures of quality of life in Britain: Some developments and trends (Hall 1976) (Social Trends No 7, HMSO, 1976)

 

I am not, and never was, a statistician, but it was clear to me from inspection of the raw data that some respondents were not using the full range of points eg some were using 0,5 and 10, others 2 thru 8 and others 5 thru 10 and so on.  I used exactly Jim Marks' syntax to create counts for all the points and then attempted to ipsatize the scores by calculating standardised scores for each individual based on their use of the scale.  The syntax for this is long lost, but the late Frank Andrews (ISR, Ann Arbor) was quite impressed and Dr Aubrey McKennell claimed that the ipsatized scores behaved in more or less the same way as the original ratings (as did the satisfaction ratings when weighted by the respondents' own importance ratings).

 

One thing that variance won't pick up is the pattern of usage by each respondent.  I went one stage further and did a count on the counts to see how many different points were used: it was a long time ago, but it must have been something like:

 

count zeros = <varlist1, varlist2 etc> (0).

count eights = ...

count nines =...

count tens= ...

count points = zeros to tens (1 thru hi)

 

  .. or in this example:

 

count points = ones to sevens (1 thru 20).

 

Another finding was that respondents using point 2 on a predictor subdomain were often scoring lower on the related main domain than respondents using 0, so Jim Ring wrote a special program to force MCA to retain the original order in modelling our findings.  There's an account of the latter in Appendix D Hall & Ring 1974 (figures and appendices) of our paper Hall & Ring 1974: Indicators of Environmental Quality and Life-Satisfaction: a subjective approach (International Sociological Association, Toronto 1974)

 

----- Original Message -----

From: [hidden email]

To: [hidden email]

Sent: Thursday, January 07, 2010 11:34 PM

Subject: Re: sytax for no variance in survey item responding

Taylor:

 

The easiest way would be use the count function.

 

not tested.

 

**find out how many answers in var1 to var20 have a value of 7 thru 1.

COUNT sevens = var1 TO var20 (7)

   /sixes = var1 TO var20 (6)

   /fives= var1 TO var20 (5)

   /fours = var1 TO var20 (4)

   /threes = var1 TO var20 (3)

   /twos = var1 TO var20 (2)

   /ones = var1 TO var20 (1)

.

 

** mark a case as suspicious if 15 or more answers are the same.

COMPUTE suspicious = ANY(15, sevens TO ones).

 

EXECUTE.

 

Note that this syntax uses the TO convention to specify all the variables between var1 and var20. You might need to list out the variables to be checked if they are not contiguous in the dataset.

 

Jim Marks

Director, Market Research

x1616

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Poling, Taylor L
Sent: Thursday, January 07, 2010 3:07 PM
To: [hidden email]
Subject: sytax for no variance in survey item responding

 

 

Hi List,

 

Can anyone suggest some syntax for flagging cases where responses are all the same across a set of many variables(i.e., representing survey items) without having to write a giant, complex IF statement?

 

Thanks,

Taylor

[Art Kendall]

compute flag = sd(var1 to var81) eq 0.


Art Kendall
Social Research Consultants

Poling, Taylor L wrote:

 

Hi List,

 

Can anyone suggest some syntax for flagging cases where responses are all the same across a set of many variables(i.e., representing survey items) without having to write a giant, complex IF statement?

 

Thanks,

Taylor

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

[Poling, Taylor Leigh]

Thanks all for the responses. I was making this much more complex than it needed to be! Computing the SD or the min value = max value solution was exactly what I was looking for. Thanks.

---

From: SPSSX(r) Discussion on behalf of Gene Maguin
Sent: Thu 1/7/2010 4:41 PM
To: [hidden email]
Subject: Re: sytax for no variance in survey item responding

Taylor,

Look at the SD.n function as in

Compute y=sd.1(x1 to x367).

If y computes to 0.00, all your variables for some case have the same value.



Gene Maguin


>>Can anyone suggest some syntax for flagging cases where responses are all
the same across a set of many variables(i.e., representing survey items)
without having to write a giant, complex IF statement?

Thanks,
Taylor

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