Generally, logistic regression

Marking criteria

Marks are shown for each question.

Answers to each question must include the syntax used and the results from that syntax.  The latter must include those items, e.g. tables, graphs, and written text, and only those items,which are appropriate to answer the questions. Assignment answers must be concise and well-presented, up to the standard of a presentation for publication.

Do not include irrelevant computer output, do ensure that appropriate information only is presented and in a logical sequence, and avoid use of excessive decimal places. Text must be clearly expressed and free from grammatical errors. Word lengths must be complied with. A ‘brief paragraph’ has a limit of 50 words and is text only.  A ‘brief report’ has a limit of 300 words, but may include one summary table and/or one figure, only where relevant.Note that paragraphs and reports must be self-contained, i.e. the main numerical findings must be presented and quantified in standard ways – follow examples given in course notes.

Marks will be deducted if these criteria are not complied with.

1.    Based on findings relating foetal exposure to tobacco in pregnancy and subsequent behaviour problems arising in early childhood, and biologically plausible explanations for this association being causal (Williams et al 1998), it is suggested that smoking in pregnancy may lead to later behavioural and developmental problems.

Use your individual dataset to examine the hypothesis that smoking in pregnancy is associated with later educational attainment of the child (at age 14), using

(a) descriptive statistics;
(b) the appropriate test of association; and
(c) a logistic regression model, with appropriate output.

Write a brief paragraph stating your conclusions.
[6 marks]

2.    Examine the relationship identified in 1 for confounding by the following socio-demographic factors (one by one):

a.    The sex of the child
b.    Family incomeduring pregnancy
c.    Mother’s education

For each of a, b and c separately, write a brief paragraph identifying whether confounding occurred, the direction and extent of confounding, and explaining why this occurred.
[6 marks]

3.    Add the variables in 2 to the logistic regression model in 1, and write a brief paragraph summarising results, including all essential information and justification, and commenting on the direction and extent of any confounding identified.
[3 marks]

4.    Examine the association between educational attainmentand aggressive behaviour at age 14 years.  Write a brief paragraphsummarising results of this analysis.
[3 marks]

5.    Re-do the analysis in question 3, adding the aggression variable to the model.Write a brief paragraph summarising results, and commenting on the appropriateness of including aggressive behaviour in the model, any differences to your model and why these changes (from the model in question 3) occurred, or if there was little change, why this occurred.
[3 marks]

6.    A plausible hypothesis is that higher maternal education might offset any adverse effect of maternal smoking in pregnancy on educational attainment.  Prepare an appropriate table (not using logistic regression) to examine this hypothesis, combining the first two categories of maternal education.
[3 marks]

7.    Fit a regression model which addresses question 6, construct an appropriate table of results, and write a brief paragraph describing the results of the analyses in questions 6 and 7.  Ignore other variables for this analysis.
[5 marks]

8.    What further analyses would you do to try to strengthen the argument for a causal effect between smoking in pregnancy specifically and educational attainment of the child?  Discuss particularly the issues around over-controlling versus adjusting for confounding (brief paragraph).  Do not do any further analysis.
[4 marks]