Dishonesty in the Australian Bureau of Statistics (eBook)

Chapter 4

Logistic Regression

What is logistic regression?

I think this question can best be answered by reference to the Population Mobility Survey which I was asked to analyse.

It was required to examine the relationship between INTENTION TO MOVE (the response variable) and the following explanatory variables:

• qualifications (QUAL) (i.e., whether graduate, post-secondary or secondary)
• whether have moved in the last 3 years (HAVMOVED)
• whether living in a separate house (SEPHOUSE)
• whether own current home (OWNHOME)
• age groups (AGECATEG)
• household type (HHOLDTYP)
• sex (SEX)

In logistic regression we aim to measure the strength of the relationship between the response variable INTENTION TO MOVE and the explanatory variables.

The test project and why I chose it

Michael Meagher, who jointly with Dina Neiger supervised me in the early stages of the project, asked me if I had a suggestion for the project I was to work on. I replied that I would like to work on a logistic regression analysis of the Population Mobility Survey. Why did I choose this project?”

I chose logistic regression because I had recently been introduced to this technique and found it both challenging and interesting. I had already carried out a logistic regression analysis of the survey on SAFETY IN THE HOME – this survey having been conducted by the ABS Victorian Office for the year 1998.

Note however that I had not studied logistic regression in my degree as it was not taught when I studied Theory of Statistics at Melbourne University in the mid-1950s.

And I chose the Population Mobility Survey as the vehicle for the logistic regression analysis because this survey was conducted by ABS Victoria in October 1999, the most recent survey conducted.

I welcomed also the opportunity to do something which had not been done in either of the previous two surveys conducted by ABS Victoria i.e., to study the effects of interaction between pairs of explanatory variables. Interaction had not been considered either in the survey GRADUATE OUTCOMES, TECHNICAL AND FURTHER EDUCATION, AUSTRALIA, 1995 or in the survey SAFETY IN THE HOME, 1998.

I mentioned the project to Neil Mclean who was my supervisor prior to Dina Neiger taking over and he was forthright “it is just a step to get you out of the ABS.” But I was confident that I would pass the test and I looked forward to the opportunity. However, I did not know at that stage what Neil Mclean knew i.e., that Senior Management had told Neil that I was going to be gotten out of the organisation one way or the other.

I forwarded a copy of my analysis of the survey SAFETY IN THE HOME to Bruce Fraser, an Assistant Director with a mathematical/statistical background, on 30 August 1999, with a request for comment.

Bruce Fraser replied on 1 September 1999 and stated “Your write-up looks a fine piece of work to me. I have a few comments, nothing too earth -shattering.”

I replied to Bruce Fraser thanking him and advising that I would take his comments on board.

My correspondence on the logistic regression analysis of the survey SAFETY IN THE HOME now follows. Note the concluding remark of Bruce Fraser:

“Thanks, Robin. No further comments.”

Schedule prepared by Michael Meagher

A meeting was held on 12 October 2000 with Dina Neiger and Michael Meagher and the following is a note on that meeting:

Additional points discussed today at meeting between Dina Neiger, Michael Meagher, Robin Rothfield

1. The tables of output from the ABS Crime and Safety Survey illustrates the type of output which may be prepared but there is no requirement to re-produce this type of output. Output should include description in “plain language” of the analysis results suitable for inclusion in the conference paper for the targeted audience of non- statisticians.
2. It is Robin’s responsibility to determine the appropriate methodology and requirements for various measures, therefore the requirements are based on the appropriateness of the methodology. Once the methodology is developed Robin will be in the position to determine whether unadjusted/adjusted ratios are appropriate.
3. Neil aims to produce a paper of about 20 pages and 6 pages of which will be based on your output, table and explanatory text. In addition, I will write a technical appendix.
4. Explanatory variables: add in occupation and sex (if data available)
5. Look at the hidden relations between the explanatory variables and the decision to move.
6. A proposal was also made that the project be supplemented by an analysis of the reasons why people move. However as there is already a table of this in the publication (table 15 of cat no. 3237.2) I wonder whether this is necessary? As mentioned in the discussion, the table in the publication is a univariate analysis. The outcome of this project should look at whether there are interactions between the explanatory variables and this variable. eg: do younger people move mainly move to be with friends and to be independent?
7. The project should be prioritised above my other activities i.e. The section’s quarterly report and preparation for the BSA course. However, it is understood that these activities are also important and it is expected that all specified activities will be completed within the specified deadlines with Robin managing conflicting priorities and timing. It was also agreed that activities specified represent a reasonable workload for someone at Robin’s level and experience, taking into account part time nature of Robin’s position.

Agreement to meet weekly with the next meeting set for 10:00 am the following Thursday.

Agreement for Robin to present a project plan with the detail and some detail of the directions he is planning to take to Michael by Monday.

Note the statement by Michael in paragraph 6:

“Agreement to meet weekly with the next meeting set for 10.00 am the following Thursday,”

The following Thursday was 19 October and there was no meeting on that date. There was also no meeting on 26 October or 2 November. The next meeting was on Friday 3 November with Dina Neiger. Moreover, there was no meeting with Michael Meagher on Monday 16 November.

I followed up on the schedule with the following email to Dina Neiger on 25 October 2000:

sex

household type (6 categories but using hhtyp1 as baseline) current home ownership (whether current home a separate house)

Noted sex not significant.

Dina responded to this with the following:

Robin

As I advised you earlier, I will be unable to look at the project (or advise you re your progress) until 3 November. Michael and I clearly outlined the work you needed to do in the two weeks between the last and the next meetings. It is up to you to make sure that the work gets done and is done appropriately.

Dina

“It is up to you to make sure that the work gets done and is done appropriately.”

Compare this with:

The job holder needs to be given clear guidance on areas where performance is not satisfactory and assistance provided, through planned learning and/or coaching, to reach a satisfactory standard….’

(ABS Performance Management Scheme Guidelines – paragraph 15.1)

On 2 November 2000 I forwarded to Dina Neiger a copy of the tasks undertaken from 20- 31 October 2000

Tasks undertaken 20-31 October 2000

Intention to move against explanatory variables - main effects

1. Carried out cross tabulations of INTENTION TO MOVE against moved (whether moved in last 3 years}, qualifications, age category, sex, household type, type of dwelling and current home ownership.) Noted a distinct relationship between INTENTION TO MOVE and qualifications, age category, type of dwelling and current home ownership. In the case of age category, it was noted that intention to move declines markedly with ascending age category.
2. Created indicative variables for qualifications, age category and household type.
3. Ran the logistic regression with INTENTION TO MOVE as response variable (i.e., intention to move in the next 3 years) and the following explanatory variables: moved (whether moved in the last 3 years) qualifications (whether graduate, other post-secondary, secondary only, but using graduate as baseline) age category (5 categories i.e. 15-24, 25-34, 35-44, 45-64, 65+, but using the youngest age category as baseline) sex household type (6 categories but using hhtyp1as baseline) current home ownership (whether current home owned) type of dwelling (whether current home a separate house).
4. Noted sex not significant.
5. Re-ran the logistic regression without sex. Noted that the following 2 among the 5- hholdtype indicative variables significant, the rest not significant:
6. hholdtype 3 (married couple with children) hholdtyp 4 (single parent with children) hholdtype 4 (single parent with ...