Watt, G. & Ecob, R. Mortality in Glasgow and Edinburgh: a paradigm of inequality in health. J. Epidemiol. Comm. Health. 1992; 46: 498-505.

Data on mortality by cause, age and sex was assembled for both Glasgow and Edinburgh for the period 1931 to 1981 from Census and death registration data. Age and sex specific mortality rates declined steadily in Glasgow and Edinburgh over this period with rates always being lower in Glasgow. Data for the period 1961 to 1981 were modelled in further detail using log-linear models with Poisson errors, using synthetic age-cohort and age period models. Age-cohort models gave a better fit to the data and fitted the data without an interaction between age and cohort (though for men there was an interaction between age and city). In both sexes an interaction was found between city and cohort. Log mortality rates were found to rise linearly with age. In 1979-83 the population of Glasgow reached a given all cause mortality rate 3.9 years earlier in men and 3.6 years earlier in women. We concluded that the current 40% cross sectional difference in mortality rates between the two cities was largely determined by levels of mortality in early adulthood or before and suggested greater priority to be given to maternal and child health particularly in areas of socio-economic disadvantage. We predicted, on the basis of the model, that differences mortality between the two cities would increase (predictions from model tested against data for 1991 in Watt and Ecob (submitted)).

Ecob, R. & Jamieson, B. A multilevel analysis of interviewer effects on a health survey. In:Westlake, A., ed. Survey and Statistical Computing. Elsevier Science, 1992: 255-68.

The twenty-07 study allocated interviewers to participants in a constrained random fashion to 'pools' of interviewers interviewing in disjoint areas (using the method of interpenetrating samples, Mahanolobis, 1946). This enabled between interviewer variability to be assessed. Gender, social class and religious affiliation of respondent were controlled for. The effect of interviewer, if any, was also allowed to vary in turn by each of these variables. The data for this paper was from the second interviewed wave (1990 aged 18) of the youngest cohort of the West of Scotland Twenty-07 study. Each interviewer was given a personality test (Cattel 16PF personality inventory) and effects of interviewer related to seven underlying factors of this inventory). Thirty measures, of health, function, self ratings - from scale - of anxiety and depression, attitudes, to life and health were investigated as well as factual items (whether steady boy/girl friend, own car etc) and interviewer ratings. Using multilevel models (using VARCL, Longford, 1990) effects of interviewer were found, for one of more sex, on the majority (22) of the 30 measures. Interviewer effects were largest for interviewer ratings and least for self perceptions. Some variations in interviewer effect ( according to social class, religion and gender) were found. Effects of interviewer were related to some aspects of their personality but not to socio-demographic or other performance characteristics (proportion of refusals, number of interviews completed). For respiratory function a more detailed model is given of interviewer effect, over both sexes simultaneously, with dummy variables corresponding to 'pool' identifiers, and allowing both interviewer variation and variance at the individual level to vary by gender of respondent.