CHAPTER 2: OBJECTIVES & METHODS
2.3 Methods
2.3.4 Data collection, management and analysis
management and analysis.
2010 and February 2011 in four urban areas of Kunming.
A face-to-face semi-structured questionnaire developed in Chinese language was adapted from FSWs and drug user behavioural surveillance developed by the Chinese National CDC16 and previous research from Kenya.17,18 The questionnaire was pretested among 24 women to determine that the content and language were appropriate for the study population and took about 50 minutes to complete. Information was collected on socio-demographic characteristics, entry into sex work, sexual behaviours, prevention knowledge and practices, substance use, reported STI symptoms, family planning and reproductive history, and health service utilization.
Double data entry was undertaken using SPSS (version 10.0) and following data checking and cleaning.
For Article 1, statistical tests were performed using Stata version 10.0 (StataCorp, College Station, Texas, U.S.A). Socio-demographic and economic characteristics of adolescent FSWs were initially presented using descriptive techniques. Chi-square tests, Wilcoxon rank sum test, fisher’s exact test and t test were used to detect differences regarding these characteristics between lower-risk and higher-risk workplaces respectively. Logistic regression was employed to investigate the extent of the association between the dependent variables and exploratory variables.
For Article 2, descriptive analysis was employed to characterise the participants. By means of the Wilcoxon rank sum test and t test we assessed associations between socio-demographic variables and the use of modern contraceptives. Logistic regression analysis was applied to determine the socio-demographic and behavioural characteristics most strongly associated with induced abortion.
Correlation of the dependent variable (prior abortion) with hypothesised factors was expressed as odds ratios (ORs) and assessed by binary logistic regression. Factors significant at the p<0.1 level in bivariate analysis were selected for inclusion in the initial multivariate logistic regression model. Stepwise forward logistic regression then
added variables starting from the variable with the lowest p value. Variables were considered significant with p<0.05. The Akaike information criterion (AIC) has been used as a measure of goodness-of-fit during the model-section procedure, the multivariate model with the smaller value of AIC was considered as the final model. To adjust for clustering due to the sampling procedures, the Stata survey option (svy) was used with 39 chains as primary sampling units. Survey-adjusted odds ratios are presented for the initial bivariate model and the final multivariate model, respectively.
ii. Cross-sectional study among adolescent FSWs conducted between July and September 2012 in four urban areas of Kunming.
A semi-structured questionnaire in Chinese was adapted from a WHO survey questionnaire for young people, the Demographic and Health Survey (DHS) youth questionnaire and a FSWs cohort study from Kenya.17,18 Key informants including leaders of FSWs’ support organizations, senior peer educators and health workers were invited to review each revision of the questionnaire. The questionnaire was pretested among 14 young FSWs to ensure the content and language were appropriate for the study population. The questionnaire covered five domains: socio-demographic information; sources of family planning information, knowledge of and experience with contraception; experience of violence; substance use; self-reported symptoms of RTI; previous pregnancy and outcomes; and health seeking behaviour.
Interviews were administrated face-to-face at different entertainment venues (e.g., karaoke, night club, dancing hall, disco, bar or personal service sectors (e.g., hair washing rooms, hair salons, massage parlour, sauna, restaurant and hotel) where young FSWs were working. Where possible, initial permission was sought from managers or owners of entertainment establishments, and an appointment arranged for a visiting time for the purposes of recruitment, while these gatekeepers also introduced the team to potentially eligible women. When requested by participants, some interviews were performed in drop-in centres within
FSWs’ support organizations. Interviews took between 40 and 50 minutes to complete.
The paper-based data was double entered using EpiData (version 3.1) by trained staff. Following data checking and cleaning, the final dataset was available for analysis. The electronic data was password protected and can only be accessed by research team.
Statistical analysis for Article 3 was conducted in Stata version 11. Univariable logistic regression was conducted to determine correlates of low SRH knowledge, unmet need for modern contraception and experience of physical and/or sexual violence. All independent variables associated with outcome variables at p<0.10 in univariable analysis were subsequently included in a multivariable logistic regression model; associations were considered significant in the multivariable model at p<0.05. The AIC has been used as a measure of goodness-of-fit during the model-section procedure, and the multivariate model with the smaller value of AIC was considered as the final model.
iii. A qualitative research project amongst adolescent girls aged 19 years or younger who reported a history of selling or exchanging sex and ATS use between March and July 2011.
The group discussions were facilitated by a semi-structured interview guide, in which participants were encouraged to discuss the relevant events and emotions, behaviours and knowledge. Core topics discussed included: participants’ background, perspectives and practices of drug use; sexual and contraceptive practices and health outcomes; health seeking behaviours; self-perception and future aspirations. Discussions started with broad personal topics then evolved to more sensitive topics in a permissive and non-judgemental way.
A typical session lasted around 1.5 hours. All discussion sessions were moderated by the lead author together with one of two trained note-takers, depending on their availability. All focus groups were digitally audio-recorded, transcribed in Chinese Mandarin, and proofread by note takers. Chinese transcripts were read through multiple times,
summarised and then thematically coded. Twelve primary themes and 26 sub-themes were identified. Coding was cross-checked by other authors fluent in Chinese.