Verbatim remarks culled from student responses in previous years:

For tables of parameter estimates and standard errors:
	(i) ALWAYS use informative labels for factor levels, for example
	white/non-white or W/NW rather than race(1) and race(2),
	male/female or M/F rather than sex(1) and sex(2).
	Avoid M/F if there is any risk that it might be confused with mother/father!

	(ii) To avoid confusion, the reference level should ordinarily be
	included in the list of coefficients for a factor parameterized by
	levels.  However, if there are only two levels, you could give the
	single value labelled sex(f) or simply female

	(iii) If the factor levels have a natural order, make sure that
	the levels are listed in that order.  Note that R may use alphabetic
	order, which may be jumbled.

Where possible, use graphs or tables to illustrate the nature of
interactions, for example a boxplot of the observed incidence rates against
dust level, split by M/F

Phrases such as "The data indicate that sex(2)s remain healthier than sex(1)s..."
should be replaced with
"The data indicate that women remain healthier than men..."

Phrases such as "the effect of a unit change in sex is to increase the
odds..." or "if the sex is changed from male to female, the odds are
increased by..." should be replaced with
"the log odds of byssinosis is xxx units higher for women than men..."
or "the odds of byssinosis are higher for women than men by a factor of 1.27".

A sentence such as "if the length of employment is changed from level 3
to level 1, the odds of being affected are decreased by..."
must be re-phrased to avoid the suggestion that the level can be changed
by administrative fiat.

Phrases such as "the odds of female is 29% higher than that of male"
and "the odds of long length of employment is 49% higher than short length"
should be replaced with
"the odds of byssinosis is 29% higher for females than for males"
and "the odds of byssinosis is higher among long-term workers than among
recent recruits by a substantial factor, estimated as exp(0.75)= 2.12."

A phrase such as "dust level 2 has a positive impact on the probability of
disease.." is ambiguous and needs to be re-thought.

English can be a difficult language to master.  The phrase,
"sex is more likely to make female workers affected by byssinosis"
means literally that sexual intercourse increases the risk of byssinosis
for women, presumably more so than for men.

The accepted term is goodness-of-fit, not good-of-fitness.

When testing for interactions, bear in mind that if 10 interactions are
tested, it is not improbable that one of the $p$-values will be less than
0.05 even if there is in fact no interaction.  Even so, it is probably unwise
to set a very stringent Bonferroni level and to look only at interactions
achieving that level of significance.  Interactions that are suspicious,
even if not significant by the Bonferroni criterion, should be checked even
if the evidence that they are non-zero is not regarded as strong.