I think we're talking at cross-purposes here. This particular sample is 100% of the population anyway, so statistical inference doesn't really come into it, unless comparisons are sought with other studies from which some questions were replicated.

At this stage of the course, my students (many of whom had little or no previous experience of statistics or computing) were still learning the language of survey research and grappling with computer technology and the mechanics of data handling. They were much happier starting from initial research questions:

Is there such a thing as "sexism"? How do we measure it? Has anyone else measured it before us?

...and only later moved to questions of methodology and/or statistical inference:

Is our measure valid? Is our measure reliable? Are there differences between groups in levels of "sexism" ? Are the differences real or could they arise by chance? How do we account for variations in levels of "sexism" ? What variables (singly or in combination) are likely to affect levels of "sexism", and to what extent?

Tutorials and exercises using this small survey (230 variables, N=142) as a teaching aid, together with others using data from the fun questionnaire completed by students at the beginning of the course (21 variables, N = 169 in waves of 25 or so per semester) helped to keep students motivated and get them up to speed with computer technology, survey research materials and routine SPSS usage. In subsequent class and homework exercises, and for the independent analysis and reporting for their chosen assessment topic, they were able to tackle, with confidence and minimal technical assistance, data from a major national survey in the public domain http://www.data-archive.ac.uk/findingData/snDescription.asp?sn=2723 (British Social Attitudes 1989, 1000+ variables, N= 3,000+)

By the end of the course we had actually covered a great deal of inferential statistics specifically geared to, and in sequence with, the analyses peformed and survey data sets used. Notes written by Jim Ring Survey Analysis Workshop - Statistical Notes (63 pages, 395kb) derive from teaching the statistical component of professional practice oriented courses in survey data collection, data management, computer processing and statistical analysis to social science students at the Polytechnic of North London from 1976 to 1992. They represented an attempt to fill a gap in the textbook provision for students who found computers and statistics daunting, and were mostly written before the appearance of the original and still much sought-after SPSS Guide to Data Analysis (Norusis,1990: later editions relating to SPSS for Windows are less helpful for beginners). They were not intended as a replacement, and should be used in conjuction with the recommended texts.

----- Original Message ----- From: ViAnn Beadle To: 'John F Hall' Sent: Tuesday, September 29, 2009 11:40 PM Subject: RE: Teaching survey analysis with SPSS

I guess I think that the survey process is about using statistics to make inferences for what is essentially categorical data. How can you make any inferences from the sample to the unmeasured population unless you're willing to look at the statistics.

From: John F Hall [mailto:johnfhall@orange.fr] Sent: Tuesday, September 29, 2009 3:13 PM To: ViAnn Beadle Cc: spssx-l@listserv.uga.edu Subject: Re: Teaching survey analysis with SPSS

This is a course for beginners, so reliabilty test way ahead of anything they need to know. This is about the survey process, not about statistics, nor for that matter about SPSS. Any software will do, but SPSS is what we used.

Items were initially selected by factor and reliability long before any preparation of tutorials, but students don't need to know that. In class I used anyway to tell them I'd cheated a bit.

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

From: ViAnn Beadle

To: 'John F Hall'

Sent: Tuesday, September 29, 2009 10:09 PM

Subject: RE: Teaching survey analysis with SPSS

No you have not missed anything here-there is no dialog box UI for the COUNT command. Note, however, if you are going to show your users how to use the dialog boxes generate syntax be sure and talk about the Execute command (what it is and why it's used) and how to turn it off when pasting transformations. BTW, do you actually run a reliability analysis on these items before glomming them into a one scale? That might be a useful exercise.

From: John F Hall [mailto:johnfhall@orange.fr] Sent: Tuesday, September 29, 2009 12:59 PM To: ViAnn Beadle; spssx-l@listserv.uga.edu Subject: Re: Teaching survey analysis with SPSS

ViAnn

I have been re-writing tutorials introducing beginners to the use of COUNT and COMPUTE to generate (admittedly crude) indices of negative attitudes to women i.e. "sexism" on data from the saved file Fifth Form Survey (questionnaire extract below)

In the tutorial I state:

It is possible to construct a crude index of "Sexism" from nine of these items, five of which are negative (a,e,f,j,o) and four positive (d,h,m,n ). To be a sexist a pupil must agree with the negative items (a,e,f,j and o) and disagree with the positive items (d,h,m and n). Thus we need to construct an index which counts the number of agreements (codes 3 and 4) with the first set together with the number of disagreements (codes 1 and 2) with the second. This will yield a score in the range 0 to 9 where 9 indicates high sexism.

The tutorial demonstrates how to do this in both direct syntax and using drop-down menus. Data are from record 2 and have varnames indicating record and field.

The syntax command:

count SEXISM=V248 V252 V253 V256 V261 (3,4) V251 V255 V259 V260 (1,2).

works perfectly and (if printed up) uses hardly any ink or paper.

However, when I try to do this using the drop down menus, I have not yet found a way to generate this index in a single pass using two different source lists, each with different values. As far as I can make out, it is necessary to create two intermediate variables, one to count agreement with items a, e, f, j and o and another to count disagreement with items d, j, m, and n and then add the two together. This process is incredibly cumbersome, and clearly demonstrates the superiority of syntax over menus for this kind of analysis.

The addition of menus has added several unnecessary pages (ink, paper?) to the tutorials, far too much to reproduce here.

The syntax generated by SPSS from the menus is:

COUNT

sexneg = v248 v252 v253 v256 v261 (3) v248 v252 v253 v256 v261 (4) .

VARIABLE LABELS sexneg 'q33 Agree a,e,f,j,o' .

EXECUTE .

COUNT

sexpos = v251 v255 v259 v260 (1) v251 v255 v259 v260 (2) .

VARIABLE LABELS sexpos 'q33 Disagree d,h,m,n' .

EXECUTE .

COMPUTE sexism = sexneg + sexpos .

EXECUTE .

Have I missed something?

My students would have given up and gone home long before they got through the menus.

John

PS It's nearly as bad using RECODE and COMPUTE to generate a superior measure which takes into account disagreement as well as agreement, and also the level of each.