1. Database
Data obtained in the interviews and ratings derived from these data were transcribed onto data sheets partly by the present author and entirely under her direction. These data sheets were submitted to the University of London Computer Centre for transfer into a computer readable form. Data files were stored as SPSS 'savefiles' and analyses carried out using the mainframe computer at ULCC.
At a later stage of the analysis the data were transferred onto the author's PC
2. Plan of analysis
Frequencies of all variables were run and scrutinised for errors and logically inconsistent values by the author and her colleague RD. Following data cleaning and consitency checks the stages o f the analysis followed the pattern of the data collection
The first stage of the analysis was divided into two parts. Initially descriptive analyses were carried out and subsequently exploratory analyses investigating the associations between these screening interview variables and the girl's self-reported depressed or anxious mood. To do this, the scores from the Great Ormond Street mood Questionnaire (GOSQ) were recoded to differentiate those with scores above or below the pre-determined cut-off point of 24/25 Associations between some of the key variables from the screening interview were also investigated.
T h e second s ta g e o f th e analysis fo llo w ed th e sa m e p a tte rn as th e first, with d e sc rip tiv e analysis
p r e c e d i n g e x p lo r a to r y analysis, and th e testin g o f th e h y p o th e s e s In th e e x p lo ra to ry stage
c o m p a r is o n s w e r e u n d e r ta k e n o f th e ca ses and c o n tro ls w h o had p a rtic ip a te d in the sec o n d
in te rv ie w . In v e s tig a tio n s w e re carried o u t to d ete rm in e th e asso c ia tio n o f 'caseness' with a
in Chapter 4.
3. Statistical tests
Data were analyzed using the Statistical Package for the Social Sciences (SPSS) initially using version X, and subsequently using SPSS/PC+.
A variety of statistical procedures were used in the analyses which are presented in the following three chapters.
The chi-square test o f significance was used to test associations between categorical data (eg, maternal marital status and 'caseness'). Throughout the presentation of results, the convention has been followed of using "significance" to describe as statistically significant those associations with a probability of .05 or less. Continuity corrections have been used where appropriate: the 'normal convention' is that such a corection should be applied when the expected cell size is less than 5 [ie, the sample size/(n. of rows x n, of columns in the table)=<5], rather than any observed values being less than S.' This included the univariate tests of association between girls with high and low scores on the GOSQ and other categorical data derived from the screening interviews. Tests of agreement between the girls and their key relatives were carried out using kappa, which measures agreement between two raters above and beyond chance agreement. The kappa statistic has a range of 0 to 1, and is usefully interpreted as:
<.01 - poor; .01-2.0 - slight; .2 1 -4 0 -fair; .41-.60 - moderate; .61-.80 - substantial; and
.81-1.00 - almost perfect (Landis & Koch, 1977).
To test the association of family rule systems with the ages of the girls, the Kruskall-Wallis one way anova (corrected for ties) was used. This test allows comparison between one variable grouped by categories of another variable (in this case the different types of parental 'rules'.
advice and teenage autonomy grouped by age), when the latter are categorised on an ordinal scale.
Multivariate analysis of variance (MANOVA) was used to establish the levels of significant association between, on the one hand, caseness, and on the other the amount of time the girls spent away from their mothers, and the numbers of admissions to hospital, both of which were recorded as continuous variables.
Because of the relatively small sample size at the intensive interview stage, analyses were carried out in two phases. Initially, first order interactions between the dependent variable (depressive or anxiety disorders) and other variables were each tested by means of chi-square. Only those associations relating to the initial hypotheses which were statistically significant were entered into subsequent multivariate analyses.
For the second stage of the study clinicians' ratings of case/not case status were the main dependent variable. This decision was taken because the purpose of the study was to look at variables associated with clinical depressive or anxiety disorder, rather than at different levels of severity of symptoms. The GOSQ, which would have provided severity ratings of self- reported symptoms, and the Total Weighted Score, which provided severity ratings based on the clinical interview with the girls, were therefore not used.
Because of the nature of the data which were used used to test the hypotheses in this study (that is, binomial case/not case status) a technique was required that dealt appropriately with such data. Thus in testing the two main hypotheses, where more than one categorical variable was being tested for best fit in relation to a single dependent variable, a loglinear model was used The logit loglinear regression model was selected’. The logit loglinear model is a special case of the general loglinear model in which one or more variables are treated as dependent and the rest as independent. Typically, and in the present study, logit models use dichotomous variables This model identifies main effects and any interactions between the variables. In the current study, this model was used to test initially for the modifying effects of the girls' confiding
“'F o r a fu ll d escrip tio n o f the proced ure u sed in arrn ing at the logit translrinuation, se e p a g e s 2 VI - 2 V2 m R oth m an , K.J. ( 1 9 8 6 ) M o d e m E p id em io lo g } . B o sto n , M A: Little. B row n & c o
relationships on the observed association between stressful life events and difficulties, and the dependent variable of caseness Two separate sets of tests were made for depressed cases and for anxiety cases. Subsequently the model was used to test for the effects of stressful effects on caseness, taking account o f the three maternal variables. Again separate tests were undertaken for depressed and anxiety cases.
CHAPTER 8 RESULTS FROM THE SCREENING INTERVIEW
This chapter reports the results from the data collected in the first (screening) interview. It will be recalled that, in this hour-long interview, the girls were seen with their key relatives. The demographic characteristics of the household were obtained, and the girls were asked questions about their health in the previous 12 months, their education and training and their work history. Both the informants filled in four questionnaires: the first reported their own psychiatric mood, and the second reported their attitudes to food and exercise, and their dietary habits. They also completed third person versions of the same questionnaires on one another.
The chapter has been divided into three sections. Section 1 briefly indicates the numbers of successful interviews and the numbers o f those who refused to take part; Sections 2, 3 and 4 cover the family demography and the personal characteristics of the girls and their parents. There is a brief overview of the characteristics of husbands/partners (section 5). Sections 6 and 7 describe the health information; section 8 gives data on time spent away from the primary carer and section 9 covers education and employment histories. Finally, section 10 gives details of the answers to the questionnaires.
1. The sample - attempted and successful interviews
Lists o f the eligible girls were compiled from each General Practice. A total of 783 names o f girls were sampled from the GPs’ lists, but the GPs excluded 26 girls who they defined as likely to be unable to partake fully in the interviews, because of limited intelligence or limited command of English. Of the remaining 757 eligible girls with whom interviews were attempted, a further 112 (15%) turned out to be ineligible when the interviewers made contact: they were either the wrong sex, the wrong age, had an inadequate command of English or had moved away without telling the GPs. This left 645 girls who were eligible and available for the study;