Re: Estimating Relationship of Total Sales and Advertising using Regression?

[Hans Chen]

Dear Listors

 

Could I conduct a multiple regression using Total Monthly Sales as dependant variable, The Number of Events (held from outside organization, my company just provide sites and service such as security), Advertising Dollar, Seasonal Dummies……as independent variables to estimate the contribution of our advertising department to our company？

 

Any suggestion would be appreciated.

 

 

Han Chen

[Peck, Jon]

I have posted a new module, SPSSINC TRANS, on SPSS Developer Central (www.spss.com/devcentral).  It provides an extension command that makes it easy to use existing Python functions to transform the case data without having to write any Python code except the one line that calls the function.  It also provides a dialog box interface.

 

This works with versions 17 or 18 of PASW Statistics and should make it easier for non-programmers to take advantage of the wealth of existing Python functions.  It can also be a time saver for users creating their own Python functions for this purpose.

 

You can read the details in the download material.  I’ve also written about it in my blog, insideout.spss.com.

 

This version is a beta.  Send bug reports and suggestions to me at [hidden email] or post on the blog.

 

Regards,

Jon Peck

[John F Hall]



Jon

 

Been playing around with your suggestions.  All the following syntax runs produce same output (copied from the SPSS Viewer).  I've changed the font to Arial 10pt, but I can't get rid of the added blank lines or change the double borders.  Provided I didn't overload them with the full list of options and statistics, my PNL students would have been able to follow both the syntax and the logic of these examples, but they would have freaked out on pivot tables. 

 

I never used artificial data: this example is taken from a survey done by three of my sophomore students in 1981 for their 2nd year group dissertation.

 

John Hall

---

MEANS TABLES=sexism2 BY sex BY ethnic

/CELLS MEAN COUNT.

---

MEANS sexism2

/crossbreak = sexism2 BY sex BY ethnic

/CELLS MEAN COUNT.

---

summarize tables=

sexism2

BY sex BY ethnic

/CELLS MEAN COUNT.

---

Report

sexism2

sex	ethnic	Mean	N
Boys	White	13.41	22
	Other	11.90	20
	Total	12.69	42
Girls	White	9.17	30
	Other	8.64	14
	Total	9.00	44
Total	White	10.96	52
	Other	10.56	34
	Total	10.80	86

 

 

I started over with just means:

 

Report

Mean

sex	ethnic	sexism2
Boys	White	13.41
	Other	11.90
	Total	12.69
Girls	White	9.17
	Other	8.64
	Total	9.00
Total	White	10.96
	Other	10.56
	Total	10.80

 

..and played around with pivot to get:

sexism2 * sex * ethnic

Mean

sex		ethnic
		White	Other	Total
Boys	sexism2	13.41	11.90	12.69
Girls	sexism2	9.17	8.64	9.00
Total	sexism2	10.96	10.56	10.80

 

By now my students would have lost me and probably have walked out: it's far too complicated for beginners. 

 

 

Going back to the table with both means and counts, I can produce:

 

sexism2 * sex * ethnic

sexism2

Mean

sex	ethnic
	White	Other	Total
Boys	13.41	11.90	12.69
Girls	9.17	8.64	9.00
Total	10.96	10.56	10.80

or:

 

sexism2 * sex * ethnic

sexism2

N

sex	ethnic
	White	Other	Total
Boys	22	20	42
Girls	30	14	44
Total	52	34	86

 

or this, the nearest I can get to what I want.

 

sexism2 * sex * ethnic

sexism2

sex	ethnic
	White		Other		Total
	Mean	N	Mean	N	Mean	N
Boys	13.41	22	11.90	20	12.69	42
Girls	9.17	30	8.64	14	9.00	44
Total	10.96	52	10.56	34	10.80	86

 

I really want the n in the same cell, preferably in brackets (n), as the mean for which it is the base.  Ideally I want the table flipped so that the global mean is top left. 

 

It's probably easier to set an exercise, provide a blank table and get them to fill it manually.

 

 

sex	ethnic
	White		Other		Total
	Mean	N	Mean	N	Mean	N
Boys	13.41	22	11.90	20	12.69	42
Girls	9.17	30	8.64	14	9.00	44
Total	10.96	52	10.56	34	10.80	86

 

 

Sexism Mean (n)	All	White	Other
All	(     )	(     )	(     )
Boys	(     )	(     )	(     )
Girls	(     )	(     )	(     )

 

 

Sexism Mean (n)	All	White	Other
All	10.8 (86)	11.0 (52)	10.6 (34)
Boys	12.7 (42)	13.4 (22)	11.9 (20)
Girls	9.0 (44)	9.2 (30)	8.6 (14)

 

Students can then see the effects of each independent variable in first and second order controls.  I know it's a very small sample, but at this point in my course it was the logic they needed to follow, not the statistics. 

 

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

From: [hidden email]

To: [hidden email]

Cc: [hidden email] ; [hidden email] ; [hidden email] ; [hidden email] ; [hidden email] ; [hidden email]

Sent: Friday, September 11, 2009 4:27 PM

Subject: RE: Crossbreak in MEANS: elaboration

MEANS and SUMMARIZE (or CTABLES) do the type of table you describe below with syntax like

MEANS TABLES=salary BY jobcat BY gender  /CELLS MEAN COUNT.

 

The initial table layout from MEANS and SUMMARIZE is different, but if you have discovered how to pivot a table, you can reshape it in many ways, including the layout below.  It is possible to reshape the tables into that layout automatically either with an autoscript or the SPSSINC MODIFY TABLES extension command available from SPSS Developer Central (www.spss.com/devcentral).

 

Although the CROSSBREAK subcommand is no longer documented, I see that it is still accepted, although integer mode is not used anymore.

 

The most general control for tables like this comes from the Custom Tables command CTABLES.  Here’s an example.

CTABLES

  /TABLE jobcat [C] BY salary [S][MEAN, COUNT] > gender [C]

/SLABELS POSITION=ROW VISIBLE=NO.

 

Regards,

Jon Peck

 

 

From: John F Hall [mailto:[hidden email]]
Sent: Friday, September 11, 2009 2:58 AM
To: Peck, Jon
Cc: [hidden email]; Stephen Tagg; Coxon Tony; David Muxworthy; Malcolm Williams; Peter Watson
Subject: Crossbreak in MEANS: elaboration

 

Jon

 

What happened to .../crossbreak in MEANS?  Any chance of restoring it?  I used to use this a lot to demonstrate the effects on means and % of introducing test variables.

 

Extract on MEANS from handout below (SPSS-X 4 on a Vax cluster and line-printer in 1992, hence Courier font)  Sample size is a bit small, but we used to do it on the British Social Attitudes (N > 3,000) survey as well.

 

A particularly useful feature of the procedure is the ability  to display  means  etc.  in  the cells of  a  table  formed  by  the categories of two independent variables (e.g. sexism means by sex and race).  We do this using CROSSBREAK:

 

     MEANS VARIABLES = SEXISM(0,9) V348(1,2)ETHNIC(1,2)

                /CROSSBREAK = SEXISM BY V348 BY ETHNIC

                /CELLS = MEAN COUNT

 

The output looks like this:

 

                    ETHNIC

              Mean :

             Count :  White      Black        Row     

                   :                         Total    

                   :       1  :       2  :

V348       --------:----------:----------:

                1  :     4.63 :     3.43 :     3.98

  Boys             :      19  :      23  :      42 

                  -:----------:----------:

                2  :     1.89 :     1.80 :     1.84

  Girls            :      19  :      25  :      44 

                  -:----------:----------:

     Column Total        3.26       2.58       2.88

                          38         48         86

 

 

A crafty use of CROSSBREAK in combination with RECODE allows  you to  display  percentages  instead  of  means  for  the  dependent variable.   What you do is to recode the value or values  of  the dependent  variable you are interested in to 100  and  everything else to 0.  The 'means' displayed are then percentages!  Thus:

 

     RECODE     SEXISM(2 THRU 7 = 100)(0,1 = 0)(ELSE = SYSMIS)

     MEANS VARIABLES = SEXISM (0,100) V348 (1,2) ETHNIC (1,2)

           /CROSSBREAK = SEXISM BY V348 BY ETHNIC

                /CELLS = MEAN COUNT

 

The output looks something like this:

 

                    ETHNIC

              Mean :

             Count :  White      Other        Row     

                   :                         Total     Epsilon (White/Other)

                   :       1  :       2  :

V348       --------:----------:----------:

                1  :   100.00 :    82.61 :    90.48        +17.4

  Boys             :      19  :      23  :      42 

                  -:----------:----------:

                2  :    47.37 :    44.00 :    45.45         +3.4

  Girls            :      19  :      25  :      44 

                  -:----------:----------:

     Column Total       73.68      62.50      67.44        +11.2

                          38         48         86

 

Epsilon (Boys/Girls)    +52.6       +38.6      +45.0

 

[NB: Epsilons (% difference) calculated by hand]

 

It would be even easier if there were a procedure:

 

elab sexism (2 thru 7) by sex by ethnic

    /cel per n.

 

where the values in the list would be converted to 100, others (except missing) to 0, and the resultant table would contain % of cases with specified values plus the base n on which % is calculated (different for each cell).  For teaching purposes, the first indepedent vars would normally be recoded as binary (high, low) to keep tables small and to make epsilon easier to calculate, but larger tables are sometimes useful as well.  There are some statistical models for elaboration in Moser & Kalton, but someone borrowed my copy years ago.

 

My old boss, the late Dr Mark Abrams, used to say, "If it's worth saying, you can say it in percentages." but that was before SPSS had charts.  Graham Kalton always used to swear by elaboration.  My pre-numerate and pre-computerate students used to think it funny when we got to BREAKDOWN in the days before it was replaced by  MEANS.

 

John Hall

[Peck, Jon]

I’m not entirely sure at this point what your desired output is, but there are several tools that can be used to change the layout, appearance, and cell contents of pivot tables.

 

Fonts and line separators can be controlled by creating a tablelook and setting this as the default in Edit>Options (or applying it manually ex post).

 

The desired pivot can be constructed by pivoting the table as needed.  This can be done manually, of course, using the pivot table editor, but it can be automated for a specific type of table by using an autoscript or the SPSSINC MODIFY TABLES extension command.  If this is too complex for beginners, they could just be given the autoscript or command syntax for this (if you don’t want to let them use the dialog box interface).

 

Pivot tables are not designed to hold multiple numbers – you wouldn’t try to do this in Excel cells either, I think.  But it is possible to force the N into the same cell as the mean using SPSSINC MERGE TABLES as in the following example (not sure how this will come across in the email, but it shows the count just below the mean in the same cell

 

Some of these tools are really designed for production applications in order to automate repetitive tasks.  They would be overkill for typical interactive use.

 

Regards,

Jon

 

 

 

Report
Mean
Gender		Employment Category		Current Salary
dimension1	Female	dimension2	Clerical	$ $25,003.69 N=206
			Manager	$ $47,213.50 N=10
			Total	$ $26,031.92 N=216
	Male	dimension2	Clerical	$ $31,558.15 N=157
			Custodial	$ $30,938.89 N=27
			Manager	$ $66,243.24 N=74
			Total	$ $41,441.78 N=258
	Total	dimension2	Clerical	$ $27,838.54 N=363
			Custodial	$ $30,938.89 N=27
			Manager	$ $63,977.80 N=84
			Total	$ $34,419.57 N=474

 

 

 

 

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of John F Hall
Sent: Monday, September 14, 2009 3:47 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Crossbreak in MEANS: elaboration

 

Jon

 

Been playing around with your suggestions.  All the following syntax runs produce same output (copied from the SPSS Viewer).  I've changed the font to Arial 10pt, but I can't get rid of the added blank lines or change the double borders.  Provided I didn't overload them with the full list of options and statistics, my PNL students would have been able to follow both the syntax and the logic of these examples, but they would have freaked out on pivot tables. 

 

I never used artificial data: this example is taken from a survey done by three of my sophomore students in 1981 for their 2nd year group dissertation.

 

John Hall

---

MEANS TABLES=sexism2 BY sex BY ethnic

/CELLS MEAN COUNT.

---

MEANS sexism2

/crossbreak = sexism2 BY sex BY ethnic

/CELLS MEAN COUNT.

---

summarize tables=

sexism2

BY sex BY ethnic

/CELLS MEAN COUNT.

---

Report

sexism2

sex	ethnic	Mean	N
Boys	White	13.41	22
	Other	11.90	20
	Total	12.69	42
Girls	White	9.17	30
	Other	8.64	14
	Total	9.00	44
Total	White	10.96	52
	Other	10.56	34
	Total	10.80	86

 

 

I started over with just means:

 

Report

Mean

sex	ethnic	sexism2
Boys	White	13.41
	Other	11.90
	Total	12.69
Girls	White	9.17
	Other	8.64
	Total	9.00
Total	White	10.96
	Other	10.56
	Total	10.80

 

..and played around with pivot to get:

sexism2 * sex * ethnic

Mean

sex		ethnic
		White	Other	Total
Boys	sexism2	13.41	11.90	12.69
Girls	sexism2	9.17	8.64	9.00
Total	sexism2	10.96	10.56	10.80

 

By now my students would have lost me and probably have walked out: it's far too complicated for beginners. 

 

 

Going back to the table with both means and counts, I can produce:

 

sexism2 * sex * ethnic

sexism2

Mean

sex	ethnic
	White	Other	Total
Boys	13.41	11.90	12.69
Girls	9.17	8.64	9.00
Total	10.96	10.56	10.80

or:

 

sexism2 * sex * ethnic

sexism2

N

sex	ethnic
	White	Other	Total
Boys	22	20	42
Girls	30	14	44
Total	52	34	86

 

or this, the nearest I can get to what I want.

 

sexism2 * sex * ethnic

sexism2

sex	ethnic
	White		Other		Total
	Mean	N	Mean	N	Mean	N
Boys	13.41	22	11.90	20	12.69	42
Girls	9.17	30	8.64	14	9.00	44
Total	10.96	52	10.56	34	10.80	86

 

I really want the n in the same cell, preferably in brackets (n), as the mean for which it is the base.  Ideally I want the table flipped so that the global mean is top left. 

 

It's probably easier to set an exercise, provide a blank table and get them to fill it manually.

sex	ethnic
	White		Other		Total
	Mean	N	Mean	N	Mean	N
Boys	13.41	22	11.90	20	12.69	42
Girls	9.17	30	8.64	14	9.00	44
Total	10.96	52	10.56	34	10.80	86

 

 

Sexism Mean (n)	All	White	Other
All	(     )	(     )	(     )
Boys	(     )	(     )	(     )
Girls	(     )	(     )	(     )

 

 

Sexism Mean (n)	All	White	Other
All	10.8 (86)	11.0 (52)	10.6 (34)
Boys	12.7 (42)	13.4 (22)	11.9 (20)
Girls	9.0 (44)	9.2 (30)	8.6 (14)

 

Students can then see the effects of each independent variable in first and second order controls.  I know it's a very small sample, but at this point in my course it was the logic they needed to follow, not the statistics. 

 

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

From: [hidden email]

To: [hidden email]

Cc: [hidden email] ; [hidden email] ; [hidden email] ; [hidden email] ; [hidden email] ; [hidden email]

Sent: Friday, September 11, 2009 4:27 PM

Subject: RE: Crossbreak in MEANS: elaboration

 

MEANS and SUMMARIZE (or CTABLES) do the type of table you describe below with syntax like

MEANS TABLES=salary BY jobcat BY gender  /CELLS MEAN COUNT.

 

The initial table layout from MEANS and SUMMARIZE is different, but if you have discovered how to pivot a table, you can reshape it in many ways, including the layout below.  It is possible to reshape the tables into that layout automatically either with an autoscript or the SPSSINC MODIFY TABLES extension command available from SPSS Developer Central (www.spss.com/devcentral).

 

Although the CROSSBREAK subcommand is no longer documented, I see that it is still accepted, although integer mode is not used anymore.

 

The most general control for tables like this comes from the Custom Tables command CTABLES.  Here’s an example.

CTABLES

  /TABLE jobcat [C] BY salary [S][MEAN, COUNT] > gender [C]

/SLABELS POSITION=ROW VISIBLE=NO.

 

Regards,

Jon Peck

 

 

From: John F Hall [mailto:[hidden email]]
Sent: Friday, September 11, 2009 2:58 AM
To: Peck, Jon
Cc: [hidden email]; Stephen Tagg; Coxon Tony; David Muxworthy; Malcolm Williams; Peter Watson
Subject: Crossbreak in MEANS: elaboration

 

Jon

 

What happened to .../crossbreak in MEANS?  Any chance of restoring it?  I used to use this a lot to demonstrate the effects on means and % of introducing test variables.

 

Extract on MEANS from handout below (SPSS-X 4 on a Vax cluster and line-printer in 1992, hence Courier font)  Sample size is a bit small, but we used to do it on the British Social Attitudes (N > 3,000) survey as well.

 

A particularly useful feature of the procedure is the ability  to display  means  etc.  in  the cells of  a  table  formed  by  the categories of two independent variables (e.g. sexism means by sex and race).  We do this using CROSSBREAK:

 

     MEANS VARIABLES = SEXISM(0,9) V348(1,2)ETHNIC(1,2)

                /CROSSBREAK = SEXISM BY V348 BY ETHNIC

                /CELLS = MEAN COUNT

 

The output looks like this:

 

                    ETHNIC

              Mean :

             Count :  White      Black        Row     

                   :                         Total    

                   :       1  :       2  :

V348       --------:----------:----------:

                1  :     4.63 :     3.43 :     3.98

  Boys             :      19  :      23  :      42 

                  -:----------:----------:

                2  :     1.89 :     1.80 :     1.84

  Girls            :      19  :      25  :      44 

                  -:----------:----------:

     Column Total        3.26       2.58       2.88

                          38         48         86

 

 

A crafty use of CROSSBREAK in combination with RECODE allows  you to  display  percentages  instead  of  means  for  the  dependent variable.   What you do is to recode the value or values  of  the dependent  variable you are interested in to 100  and  everything else to 0.  The 'means' displayed are then percentages!  Thus:

 

     RECODE     SEXISM(2 THRU 7 = 100)(0,1 = 0)(ELSE = SYSMIS)

     MEANS VARIABLES = SEXISM (0,100) V348 (1,2) ETHNIC (1,2)

           /CROSSBREAK = SEXISM BY V348 BY ETHNIC

                /CELLS = MEAN COUNT

 

The output looks something like this:

 

                    ETHNIC

              Mean :

             Count :  White      Other        Row     

                   :                         Total     Epsilon (White/Other)

                   :       1  :       2  :

V348       --------:----------:----------:

                1  :   100.00 :    82.61 :    90.48        +17.4

  Boys             :      19  :      23  :      42 

                  -:----------:----------:

                2  :    47.37 :    44.00 :    45.45         +3.4

  Girls            :      19  :      25  :      44 

                  -:----------:----------:

     Column Total       73.68      62.50      67.44        +11.2

                          38         48         86

 

Epsilon (Boys/Girls)    +52.6       +38.6      +45.0

 

[NB: Epsilons (% difference) calculated by hand]

 

It would be even easier if there were a procedure:

 

elab sexism (2 thru 7) by sex by ethnic

    /cel per n.

 

where the values in the list would be converted to 100, others (except missing) to 0, and the resultant table would contain % of cases with specified values plus the base n on which % is calculated (different for each cell).  For teaching purposes, the first indepedent vars would normally be recoded as binary (high, low) to keep tables small and to make epsilon easier to calculate, but larger tables are sometimes useful as well.  There are some statistical models for elaboration in Moser & Kalton, but someone borrowed my copy years ago.

 

My old boss, the late Dr Mark Abrams, used to say, "If it's worth saying, you can say it in percentages." but that was before SPSS had charts.  Graham Kalton always used to swear by elaboration.  My pre-numerate and pre-computerate students used to think it funny when we got to BREAKDOWN in the days before it was replaced by  MEANS.

 

John Hall

[John F Hall]



Jon

 

Thanks for the effort you have put in on this.

 

I've used a bigger survey this time (Quality of Life in Britain, N=903)  It's one I did myself in 1975, in collaboration with the late Angus Campbell (ISR Ann Arbor)

 

Dependent variable is mean score on 1-7 semantic differential item, Tied down....Free

 

means var561 by sex by agegroup /cel mea cou.

 

produces:

 

LIFE IS TIED DOWN OR FREE

Sex of Respondent	Agegroup of repondent	Mean	N
Men	17-29	4.91	88
	30-44	4.86	90
	45-59	5.07	109
	60 or over	5.71	90
	Total	5.14	377
Women	17-29	4.33	115
	30-44	4.44	124
	45-59	5.23	132
	60 or over	5.60	155
	Total	4.96	526
Total	17-29	4.58	203
	30-44	4.61	214
	45-59	5.16	241
	60 or over	5.64	245
	Total	5.03	903

 

..and if I drag agegroup from rows to columns, then statistics from columns to rows, I get something very close to what I'm looking for. 

LIFE IS TIED DOWN OR FREE

Sex of Respondent		Agegroup of repondent
		17-29	30-44	45-59	60 or over	Total
Men	Mean	4.91	4.86	5.07	5.71	5.14
	N	88	90	109	90	377
Women	Mean	4.33	4.44	5.23	5.60	4.96
	N	115	124	132	155	526
Total	Mean	4.58	4.61	5.16	5.64	5.03
	N	203	214	241	245	903

Even the most recalcitrant student should be able follow the mechanics and understand the output.  I don't know if you've ever been in a computer lab trying to teach a crowd of students, but when we tried it with only 8 students on an early version of SPSS on a PC (PCPlus I think) absolute chaos ensued, so we went back to using VDUs on the Vax mainframe with VMS and EDIT plus a fantastic front end written by Jim Ring, where we could effectively manage 24 students.

Anyway, this table is pretty good, especially since it has nice consistent gradients.  For teaching purposes, examples like this are sometimes difficult to find in real survey data, and the searches take up significant time.  However, for demonstrating the logic of analysis, it would be nice if the table could be flipped top left to bottom right, but keeping the means rows above the n rows.

 

		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean	5.14	4.91	4.86	5.07	5.71
	N	377	88	90	109	90
Women	Mean	4.96	4.33	4.44	5.23	5.60
	N	526	115	124	132	155

 

Manually tidied up, but I don't know how to get it perfect.  SPSS formats tables differently from Word.  Non-numerate students learn a lot from this kind of thing: it makes them think.  Colour text certainly helps.

 

		Agegroup of respondent
		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean N	5.14 377	4.91 88	4.86 90	5.07 109	5.71 90
Women	Mean N	4.96 526	4.33 115	4.44 124	5.23 132	5.60 155

 

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

From: [hidden email]

To: [hidden email] ; [hidden email]

Sent: Monday, September 14, 2009 4:35 PM

Subject: RE: Re: [SPSSX-L] Crossbreak in MEANS: elaboration

I’m not entirely sure at this point what your desired output is, but there are several tools that can be used to change the layout, appearance, and cell contents of pivot tables.

 

Fonts and line separators can be controlled by creating a tablelook and setting this as the default in Edit>Options (or applying it manually ex post).

 

The desired pivot can be constructed by pivoting the table as needed.  This can be done manually, of course, using the pivot table editor, but it can be automated for a specific type of table by using an autoscript or the SPSSINC MODIFY TABLES extension command.  If this is too complex for beginners, they could just be given the autoscript or command syntax for this (if you don’t want to let them use the dialog box interface).

 

Pivot tables are not designed to hold multiple numbers – you wouldn’t try to do this in Excel cells either, I think.  But it is possible to force the N into the same cell as the mean using SPSSINC MERGE TABLES as in the following example (not sure how this will come across in the email, but it shows the count just below the mean in the same cell

 

Some of these tools are really designed for production applications in order to automate repetitive tasks.  They would be overkill for typical interactive use.

 

Regards,

Jon

 

 

 

Report
Mean
Gender		Employment Category		Current Salary
dimension1	Female	dimension2	Clerical	$ $25,003.69 N=206
			Manager	$ $47,213.50 N=10
			Total	$ $26,031.92 N=216
	Male	dimension2	Clerical	$ $31,558.15 N=157
			Custodial	$ $30,938.89 N=27
			Manager	$ $66,243.24 N=74
			Total	$ $41,441.78 N=258
	Total	dimension2	Clerical	$ $27,838.54 N=363
			Custodial	$ $30,938.89 N=27
			Manager	$ $63,977.80 N=84
			Total	$ $34,419.57 N=474

 

 

 

 

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of John F Hall
Sent: Monday, September 14, 2009 3:47 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Crossbreak in MEANS: elaboration

 

Jon

 

Been playing around with your suggestions.  All the following syntax runs produce same output (copied from the SPSS Viewer).  I've changed the font to Arial 10pt, but I can't get rid of the added blank lines or change the double borders.  Provided I didn't overload them with the full list of options and statistics, my PNL students would have been able to follow both the syntax and the logic of these examples, but they would have freaked out on pivot tables. 

 

I never used artificial data: this example is taken from a survey done by three of my sophomore students in 1981 for their 2nd year group dissertation.

 

John Hall

---

MEANS TABLES=sexism2 BY sex BY ethnic

/CELLS MEAN COUNT.

---

MEANS sexism2

/crossbreak = sexism2 BY sex BY ethnic

/CELLS MEAN COUNT.

---

summarize tables=

sexism2

BY sex BY ethnic

/CELLS MEAN COUNT.

---

Report

sexism2

sex	ethnic	Mean	N
Boys	White	13.41	22
	Other	11.90	20
	Total	12.69	42
Girls	White	9.17	30
	Other	8.64	14
	Total	9.00	44
Total	White	10.96	52
	Other	10.56	34
	Total	10.80	86

 

 

I started over with just means:

 

Report

Mean

sex	ethnic	sexism2
Boys	White	13.41
	Other	11.90
	Total	12.69
Girls	White	9.17
	Other	8.64
	Total	9.00
Total	White	10.96
	Other	10.56
	Total	10.80

 

..and played around with pivot to get:

sexism2 * sex * ethnic

Mean

sex		ethnic
		White	Other	Total
Boys	sexism2	13.41	11.90	12.69
Girls	sexism2	9.17	8.64	9.00
Total	sexism2	10.96	10.56	10.80

 

By now my students would have lost me and probably have walked out: it's far too complicated for beginners. 

 

 

Going back to the table with both means and counts, I can produce:

 

sexism2 * sex * ethnic

sexism2

Mean

sex	ethnic
	White	Other	Total
Boys	13.41	11.90	12.69
Girls	9.17	8.64	9.00
Total	10.96	10.56	10.80

or:

 

sexism2 * sex * ethnic

sexism2

N

sex	ethnic
	White	Other	Total
Boys	22	20	42
Girls	30	14	44
Total	52	34	86

 

or this, the nearest I can get to what I want.

 

sexism2 * sex * ethnic

sexism2

sex	ethnic
	White		Other		Total
	Mean	N	Mean	N	Mean	N
Boys	13.41	22	11.90	20	12.69	42
Girls	9.17	30	8.64	14	9.00	44
Total	10.96	52	10.56	34	10.80	86

 

I really want the n in the same cell, preferably in brackets (n), as the mean for which it is the base.  Ideally I want the table flipped so that the global mean is top left. 

 

It's probably easier to set an exercise, provide a blank table and get them to fill it manually.

sex	ethnic
	White		Other		Total
	Mean	N	Mean	N	Mean	N
Boys	13.41	22	11.90	20	12.69	42
Girls	9.17	30	8.64	14	9.00	44
Total	10.96	52	10.56	34	10.80	86

 

 

Sexism Mean (n)	All	White	Other
All	(     )	(     )	(     )
Boys	(     )	(     )	(     )
Girls	(     )	(     )	(     )

 

 

Sexism Mean (n)	All	White	Other
All	10.8 (86)	11.0 (52)	10.6 (34)
Boys	12.7 (42)	13.4 (22)	11.9 (20)
Girls	9.0 (44)	9.2 (30)	8.6 (14)

 

Students can then see the effects of each independent variable in first and second order controls.  I know it's a very small sample, but at this point in my course it was the logic they needed to follow, not the statistics. 

 

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

From: [hidden email]

To: [hidden email]

Cc: [hidden email] ; [hidden email] ; [hidden email] ; [hidden email] ; [hidden email] ; [hidden email]

Sent: Friday, September 11, 2009 4:27 PM

Subject: RE: Crossbreak in MEANS: elaboration

 

MEANS and SUMMARIZE (or CTABLES) do the type of table you describe below with syntax like

MEANS TABLES=salary BY jobcat BY gender  /CELLS MEAN COUNT.

 

The initial table layout from MEANS and SUMMARIZE is different, but if you have discovered how to pivot a table, you can reshape it in many ways, including the layout below.  It is possible to reshape the tables into that layout automatically either with an autoscript or the SPSSINC MODIFY TABLES extension command available from SPSS Developer Central (www.spss.com/devcentral).

 

Although the CROSSBREAK subcommand is no longer documented, I see that it is still accepted, although integer mode is not used anymore.

 

The most general control for tables like this comes from the Custom Tables command CTABLES.  Here’s an example.

CTABLES

  /TABLE jobcat [C] BY salary [S][MEAN, COUNT] > gender [C]

/SLABELS POSITION=ROW VISIBLE=NO.

 

Regards,

Jon Peck

 

 

From: John F Hall [mailto:[hidden email]]
Sent: Friday, September 11, 2009 2:58 AM
To: Peck, Jon
Cc: [hidden email]; Stephen Tagg; Coxon Tony; David Muxworthy; Malcolm Williams; Peter Watson
Subject: Crossbreak in MEANS: elaboration

 

Jon

 

What happened to .../crossbreak in MEANS?  Any chance of restoring it?  I used to use this a lot to demonstrate the effects on means and % of introducing test variables.

 

Extract on MEANS from handout below (SPSS-X 4 on a Vax cluster and line-printer in 1992, hence Courier font)  Sample size is a bit small, but we used to do it on the British Social Attitudes (N > 3,000) survey as well.

 

A particularly useful feature of the procedure is the ability  to display  means  etc.  in  the cells of  a  table  formed  by  the categories of two independent variables (e.g. sexism means by sex and race).  We do this using CROSSBREAK:

 

     MEANS VARIABLES = SEXISM(0,9) V348(1,2)ETHNIC(1,2)

                /CROSSBREAK = SEXISM BY V348 BY ETHNIC

                /CELLS = MEAN COUNT

 

The output looks like this:

 

                    ETHNIC

              Mean :

             Count :  White      Black        Row     

                   :                         Total    

                   :       1  :       2  :

V348       --------:----------:----------:

                1  :     4.63 :     3.43 :     3.98

  Boys             :      19  :      23  :      42 

                  -:----------:----------:

                2  :     1.89 :     1.80 :     1.84

  Girls            :      19  :      25  :      44 

                  -:----------:----------:

     Column Total        3.26       2.58       2.88

                          38         48         86

 

 

A crafty use of CROSSBREAK in combination with RECODE allows  you to  display  percentages  instead  of  means  for  the  dependent variable.   What you do is to recode the value or values  of  the dependent  variable you are interested in to 100  and  everything else to 0.  The 'means' displayed are then percentages!  Thus:

 

     RECODE     SEXISM(2 THRU 7 = 100)(0,1 = 0)(ELSE = SYSMIS)

     MEANS VARIABLES = SEXISM (0,100) V348 (1,2) ETHNIC (1,2)

           /CROSSBREAK = SEXISM BY V348 BY ETHNIC

                /CELLS = MEAN COUNT

 

The output looks something like this:

 

                    ETHNIC

              Mean :

             Count :  White      Other        Row     

                   :                         Total     Epsilon (White/Other)

                   :       1  :       2  :

V348       --------:----------:----------:

                1  :   100.00 :    82.61 :    90.48        +17.4

  Boys             :      19  :      23  :      42 

                  -:----------:----------:

                2  :    47.37 :    44.00 :    45.45         +3.4

  Girls            :      19  :      25  :      44 

                  -:----------:----------:

     Column Total       73.68      62.50      67.44        +11.2

                          38         48         86

 

Epsilon (Boys/Girls)    +52.6       +38.6      +45.0

 

[NB: Epsilons (% difference) calculated by hand]

 

It would be even easier if there were a procedure:

 

elab sexism (2 thru 7) by sex by ethnic

    /cel per n.

 

where the values in the list would be converted to 100, others (except missing) to 0, and the resultant table would contain % of cases with specified values plus the base n on which % is calculated (different for each cell).  For teaching purposes, the first indepedent vars would normally be recoded as binary (high, low) to keep tables small and to make epsilon easier to calculate, but larger tables are sometimes useful as well.  There are some statistical models for elaboration in Moser & Kalton, but someone borrowed my copy years ago.

 

My old boss, the late Dr Mark Abrams, used to say, "If it's worth saying, you can say it in percentages." but that was before SPSS had charts.  Graham Kalton always used to swear by elaboration.  My pre-numerate and pre-computerate students used to think it funny when we got to BREAKDOWN in the days before it was replaced by  MEANS.

 

John Hall

[Peck, Jon]

If you want that kind of detailed control, Custom Tables is probably what you want.  It would let you put the totals at the top, order the statistics and categories, etc.

 

Here’s an example using the GSS file shipped with PASW Statistics.

 

First I binned the age variable into four equal bins using the Visual Binner.  That generates this syntax.

 

VARIABLE LABELS  agebinned 'Age of Respondent (Binned)'.

FORMATS  agebinned (F5.0).

VALUE LABELS  agebinned 1 '<= 29' 2 '30 - 37' 3 '38 - 47' 4 '48 - 64' 5 '65+' 98 'DK' 99 'NA'.

MISSING VALUES  agebinned (0, 98, 99).

VARIABLE LEVEL  agebinned (ORDINAL).

 

Then I ran CTABLES to get a table structured the way you showed.

CTABLES

  /TABLE sex BY agebinned > life [S][MEAN, COUNT]

  /SLABELS POSITION=ROW

  /CATEGORIES VARIABLES=sex ORDER=A KEY=VALUE EMPTY=INCLUDE TOTAL=YES POSITION=BEFORE

  /CATEGORIES VARIABLES=agebinned ORDER=A KEY=VALUE EMPTY=INCLUDE TOTAL=YES POSITION=AFTER

    MISSING=EXCLUDE.

 

Then, for good measure, I used SPSSINC MODIFY TABLES to make all the counts light gray so that focus would be on the means.

 

SPSSINC MODIFY TABLES subtype="'Custom Table'"

SELECT=Count

DIMENSION= ROWS

LEVEL = -1  PROCESS = PRECEDING

/STYLES  APPLYTO=BOTH

TEXTCOLOR = 200 200 200.

 

I could also have bolded the totals or, with a tiny custom function, changed the fonts, point size, etc of selected cells with MODIFY TABLES  if I didn’t want to do that manually.

 

Regards,

Jon

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of John F Hall
Sent: Monday, September 14, 2009 11:18 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Crossbreak in MEANS: elaboration

 

Jon

 

Thanks for the effort you have put in on this.

 

I've used a bigger survey this time (Quality of Life in Britain, N=903)  It's one I did myself in 1975, in collaboration with the late Angus Campbell (ISR Ann Arbor)

 

Dependent variable is mean score on 1-7 semantic differential item, Tied down....Free

 

means var561 by sex by agegroup /cel mea cou.

 

produces:

 

LIFE IS TIED DOWN OR FREE

Sex of Respondent	Agegroup of repondent	Mean	N
Men	17-29	4.91	88
	30-44	4.86	90
	45-59	5.07	109
	60 or over	5.71	90
	Total	5.14	377
Women	17-29	4.33	115
	30-44	4.44	124
	45-59	5.23	132
	60 or over	5.60	155
	Total	4.96	526
Total	17-29	4.58	203
	30-44	4.61	214
	45-59	5.16	241
	60 or over	5.64	245
	Total	5.03	903

 

..and if I drag agegroup from rows to columns, then statistics from columns to rows, I get something very close to what I'm looking for. 

LIFE IS TIED DOWN OR FREE

Sex of Respondent		Agegroup of repondent
		17-29	30-44	45-59	60 or over	Total
Men	Mean	4.91	4.86	5.07	5.71	5.14
	N	88	90	109	90	377
Women	Mean	4.33	4.44	5.23	5.60	4.96
	N	115	124	132	155	526
Total	Mean	4.58	4.61	5.16	5.64	5.03
	N	203	214	241	245	903

Even the most recalcitrant student should be able follow the mechanics and understand the output.  I don't know if you've ever been in a computer lab trying to teach a crowd of students, but when we tried it with only 8 students on an early version of SPSS on a PC (PCPlus I think) absolute chaos ensued, so we went back to using VDUs on the Vax mainframe with VMS and EDIT plus a fantastic front end written by Jim Ring, where we could effectively manage 24 students.

Anyway, this table is pretty good, especially since it has nice consistent gradients.  For teaching purposes, examples like this are sometimes difficult to find in real survey data, and the searches take up significant time.  However, for demonstrating the logic of analysis, it would be nice if the table could be flipped top left to bottom right, but keeping the means rows above the n rows.

 

		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean	5.14	4.91	4.86	5.07	5.71
	N	377	88	90	109	90
Women	Mean	4.96	4.33	4.44	5.23	5.60
	N	526	115	124	132	155

 

Manually tidied up, but I don't know how to get it perfect.  SPSS formats tables differently from Word.  Non-numerate students learn a lot from this kind of thing: it makes them think.  Colour text certainly helps.

 

		Agegroup of respondent
		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean N	5.14 377	4.91 88	4.86 90	5.07 109	5.71 90
Women	Mean N	4.96 526	4.33 115	4.44 124	5.23 132	5.60 155

 

[John F Hall]



Wouldn't it be easier just to restore the original  Means..../crossbreak facility exactly as it was in 1992?

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

From: [hidden email]

To: [hidden email]

Sent: Monday, September 14, 2009 9:08 PM

Subject: Re: Crossbreak in MEANS: elaboration

If you want that kind of detailed control, Custom Tables is probably what you want.  It would let you put the totals at the top, order the statistics and categories, etc.

 

Here’s an example using the GSS file shipped with PASW Statistics.

 

First I binned the age variable into four equal bins using the Visual Binner.  That generates this syntax.

 

VARIABLE LABELS  agebinned 'Age of Respondent (Binned)'.

FORMATS  agebinned (F5.0).

VALUE LABELS  agebinned 1 '<= 29' 2 '30 - 37' 3 '38 - 47' 4 '48 - 64' 5 '65+' 98 'DK' 99 'NA'.

MISSING VALUES  agebinned (0, 98, 99).

VARIABLE LEVEL  agebinned (ORDINAL).

 

Then I ran CTABLES to get a table structured the way you showed.

CTABLES

  /TABLE sex BY agebinned > life [S][MEAN, COUNT]

  /SLABELS POSITION=ROW

  /CATEGORIES VARIABLES=sex ORDER=A KEY=VALUE EMPTY=INCLUDE TOTAL=YES POSITION=BEFORE

  /CATEGORIES VARIABLES=agebinned ORDER=A KEY=VALUE EMPTY=INCLUDE TOTAL=YES POSITION=AFTER

    MISSING=EXCLUDE.

 

Then, for good measure, I used SPSSINC MODIFY TABLES to make all the counts light gray so that focus would be on the means.

 

SPSSINC MODIFY TABLES subtype="'Custom Table'"

SELECT=Count

DIMENSION= ROWS

LEVEL = -1  PROCESS = PRECEDING

/STYLES  APPLYTO=BOTH

TEXTCOLOR = 200 200 200.

 

I could also have bolded the totals or, with a tiny custom function, changed the fonts, point size, etc of selected cells with MODIFY TABLES  if I didn’t want to do that manually.

 

Regards,

Jon

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of John F Hall
Sent: Monday, September 14, 2009 11:18 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Crossbreak in MEANS: elaboration

 

Jon

 

Thanks for the effort you have put in on this.

 

I've used a bigger survey this time (Quality of Life in Britain, N=903)  It's one I did myself in 1975, in collaboration with the late Angus Campbell (ISR Ann Arbor)

 

Dependent variable is mean score on 1-7 semantic differential item, Tied down....Free

 

means var561 by sex by agegroup /cel mea cou.

 

produces:

 

LIFE IS TIED DOWN OR FREE

Sex of Respondent	Agegroup of repondent	Mean	N
Men	17-29	4.91	88
	30-44	4.86	90
	45-59	5.07	109
	60 or over	5.71	90
	Total	5.14	377
Women	17-29	4.33	115
	30-44	4.44	124
	45-59	5.23	132
	60 or over	5.60	155
	Total	4.96	526
Total	17-29	4.58	203
	30-44	4.61	214
	45-59	5.16	241
	60 or over	5.64	245
	Total	5.03	903

 

..and if I drag agegroup from rows to columns, then statistics from columns to rows, I get something very close to what I'm looking for. 

LIFE IS TIED DOWN OR FREE

Sex of Respondent		Agegroup of repondent
		17-29	30-44	45-59	60 or over	Total
Men	Mean	4.91	4.86	5.07	5.71	5.14
	N	88	90	109	90	377
Women	Mean	4.33	4.44	5.23	5.60	4.96
	N	115	124	132	155	526
Total	Mean	4.58	4.61	5.16	5.64	5.03
	N	203	214	241	245	903

Even the most recalcitrant student should be able follow the mechanics and understand the output.  I don't know if you've ever been in a computer lab trying to teach a crowd of students, but when we tried it with only 8 students on an early version of SPSS on a PC (PCPlus I think) absolute chaos ensued, so we went back to using VDUs on the Vax mainframe with VMS and EDIT plus a fantastic front end written by Jim Ring, where we could effectively manage 24 students.

Anyway, this table is pretty good, especially since it has nice consistent gradients.  For teaching purposes, examples like this are sometimes difficult to find in real survey data, and the searches take up significant time.  However, for demonstrating the logic of analysis, it would be nice if the table could be flipped top left to bottom right, but keeping the means rows above the n rows.

 

		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean	5.14	4.91	4.86	5.07	5.71
	N	377	88	90	109	90
Women	Mean	4.96	4.33	4.44	5.23	5.60
	N	526	115	124	132	155

 

Manually tidied up, but I don't know how to get it perfect.  SPSS formats tables differently from Word.  Non-numerate students learn a lot from this kind of thing: it makes them think.  Colour text certainly helps.

 

		Agegroup of respondent
		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean N	5.14 377	4.91 88	4.86 90	5.07 109	5.71 90
Women	Mean N	4.96 526	4.33 115	4.44 124	5.23 132	5.60 155

 

[ViAnn Beadle]

Nope. It would require completely recreating old ASCII output. That bridge was burned 13 years ago.

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of John F Hall
Sent: Monday, September 14, 2009 1:56 PM
To: [hidden email]
Subject: Re: Crossbreak in MEANS: elaboration

 

Wouldn't it be easier just to restore the original  Means..../crossbreak facility exactly as it was in 1992?

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

From: [hidden email]

To: [hidden email]

Sent: Monday, September 14, 2009 9:08 PM

Subject: Re: Crossbreak in MEANS: elaboration

 

If you want that kind of detailed control, Custom Tables is probably what you want.  It would let you put the totals at the top, order the statistics and categories, etc.

 

Here’s an example using the GSS file shipped with PASW Statistics.

 

First I binned the age variable into four equal bins using the Visual Binner.  That generates this syntax.

 

VARIABLE LABELS  agebinned 'Age of Respondent (Binned)'.

FORMATS  agebinned (F5.0).

VALUE LABELS  agebinned 1 '<= 29' 2 '30 - 37' 3 '38 - 47' 4 '48 - 64' 5 '65+' 98 'DK' 99 'NA'.

MISSING VALUES  agebinned (0, 98, 99).

VARIABLE LEVEL  agebinned (ORDINAL).

 

Then I ran CTABLES to get a table structured the way you showed.

CTABLES

  /TABLE sex BY agebinned > life [S][MEAN, COUNT]

  /SLABELS POSITION=ROW

  /CATEGORIES VARIABLES=sex ORDER=A KEY=VALUE EMPTY=INCLUDE TOTAL=YES POSITION=BEFORE

  /CATEGORIES VARIABLES=agebinned ORDER=A KEY=VALUE EMPTY=INCLUDE TOTAL=YES POSITION=AFTER

    MISSING=EXCLUDE.

 

Then, for good measure, I used SPSSINC MODIFY TABLES to make all the counts light gray so that focus would be on the means.

 

SPSSINC MODIFY TABLES subtype="'Custom Table'"

SELECT=Count

DIMENSION= ROWS

LEVEL = -1  PROCESS = PRECEDING

/STYLES  APPLYTO=BOTH

TEXTCOLOR = 200 200 200.

 

I could also have bolded the totals or, with a tiny custom function, changed the fonts, point size, etc of selected cells with MODIFY TABLES  if I didn’t want to do that manually.

 

Regards,

Jon

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of John F Hall
Sent: Monday, September 14, 2009 11:18 AM
To: [hidden email]
Subject: Re: [SPSSX-L] Crossbreak in MEANS: elaboration

 

Jon

 

Thanks for the effort you have put in on this.

 

I've used a bigger survey this time (Quality of Life in Britain, N=903)  It's one I did myself in 1975, in collaboration with the late Angus Campbell (ISR Ann Arbor)

 

Dependent variable is mean score on 1-7 semantic differential item, Tied down....Free

 

means var561 by sex by agegroup /cel mea cou.

 

produces:

 

LIFE IS TIED DOWN OR FREE

Sex of Respondent	Agegroup of repondent	Mean	N
Men	17-29	4.91	88
	30-44	4.86	90
	45-59	5.07	109
	60 or over	5.71	90
	Total	5.14	377
Women	17-29	4.33	115
	30-44	4.44	124
	45-59	5.23	132
	60 or over	5.60	155
	Total	4.96	526
Total	17-29	4.58	203
	30-44	4.61	214
	45-59	5.16	241
	60 or over	5.64	245
	Total	5.03	903

 

..and if I drag agegroup from rows to columns, then statistics from columns to rows, I get something very close to what I'm looking for. 

LIFE IS TIED DOWN OR FREE

Sex of Respondent		Agegroup of repondent
		17-29	30-44	45-59	60 or over	Total
Men	Mean	4.91	4.86	5.07	5.71	5.14
	N	88	90	109	90	377
Women	Mean	4.33	4.44	5.23	5.60	4.96
	N	115	124	132	155	526
Total	Mean	4.58	4.61	5.16	5.64	5.03
	N	203	214	241	245	903

Even the most recalcitrant student should be able follow the mechanics and understand the output.  I don't know if you've ever been in a computer lab trying to teach a crowd of students, but when we tried it with only 8 students on an early version of SPSS on a PC (PCPlus I think) absolute chaos ensued, so we went back to using VDUs on the Vax mainframe with VMS and EDIT plus a fantastic front end written by Jim Ring, where we could effectively manage 24 students.

Anyway, this table is pretty good, especially since it has nice consistent gradients.  For teaching purposes, examples like this are sometimes difficult to find in real survey data, and the searches take up significant time.  However, for demonstrating the logic of analysis, it would be nice if the table could be flipped top left to bottom right, but keeping the means rows above the n rows.

 

		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean	5.14	4.91	4.86	5.07	5.71
	N	377	88	90	109	90
Women	Mean	4.96	4.33	4.44	5.23	5.60
	N	526	115	124	132	155

 

Manually tidied up, but I don't know how to get it perfect.  SPSS formats tables differently from Word.  Non-numerate students learn a lot from this kind of thing: it makes them think.  Colour text certainly helps.

 

		Agegroup of respondent
		Total	17-29	30-44	45-59	60 or over
Total	Mean	5.03	4.58	4.61	5.16	5.64
	N	903	203	214	241	245
Men	Mean N	5.14 377	4.91 88	4.86 90	5.07 109	5.71 90
Women	Mean N	4.96 526	4.33 115	4.44 124	5.23 132	5.60 155