Chapter 3 – Methodology
3.5. Phase 1: Web-survey on food intake among university students
3.5.2 Development and refinement of the web-survey: details of the
Prior to data collection, the web-survey was tested among a sample of University of Sheffield students. Any problems associated with completion of the questionnaire and understanding of questions could therefore be detected – and acted upon - before data collection began, thus optimising success of the main study. This pilot study was also designed to estimate a response rate that could be expected in the main study, and to assess if different dietary patterns might exist within a UK university student population, thus justifying use of a dietary patterns approach to analysis.
To address these objectives, 40 students at the University of Sheffield completed the web-survey and 11 of these participated in a follow-up focus group about the survey. Students were recruited via email to reflect the sampling procedure to be used in the main study. Snowball sampling was also employed to enhance sample size. Focus groups addressed: range of response options and any ambiguities encountered when answering questions; length of the survey; use of incentives; use of promotional web-page; habitually eaten foods that were missing from the FFQ; suitability of the FFQ for international students; and any other improvements that could be made to the survey (Appendix B).
The pilot study sample comprised predominantly female (n=26) and postgraduate (n=35) students, with mean age of 25 years. A total of 50% of students indicated that they were of White British ethnicity; the remaining 50% comprised a range of ethnicities. Mean BMI was 22.7 kg/m2.Of the 11 focus group participants, four were male and seven were female. Focus group participants also embraced a range of ethnicities and stages of study.
The overall response rate was estimated at approximately 8% (excluding snowball respondents), comparable to a 5% response rate reported in an online survey assessing stress and body weight among UK undergraduate students (173). Response rate for participation in follow-up focus groups was 30% (comprising participants recruited via both sampling methods). Research employing data-driven multivariate analytical techniques such as principal components analysis requires large sample sizes, calculated using the formula ‘n ≥ 50 + 8m’, where m represents the number of dietary variables entered into the analysis (243). In line with the dietary patterns literature the number of dietary input variables in the main study was 55 (22,24,28,74), and a minimum sample size of 426 students would therefore be required. In excess of 24,000 students attended the University of Sheffield in 2012 (244) and a response rate of 8% would thus have achieved a sufficiently large sample (n = 1920 students) from this single university. However, it was decided that a multi-centre study would ensure target sample size was achieved, as well as increase the diversity of students sampled, enhancing external validity.
Approximately half of all students completing the survey indicated that it was too long; a 20-25 minute completion time was envisaged, which is longer than most online surveys that students are asked to complete. This is noteworthy, since longer surveys may compromise both response and retention rates (245). Efforts to reduce completion time, however, were restricted by the necessity to incorporate the full FFQ into the survey to optimise validity of dietary intake data obtained. Such efforts were therefore limited to the exclusion of questions within the second part of the web-survey, which had to be balanced against the desire to obtain sufficient non-dietary data to interpret the dietary patterns obtained. After consideration, the survey was kept in its current form, and participants were provided with clearer information on completion time in the main study. The author acknowledges that the length of the survey may have contributed towards a selection bias, such that students more interested in diet and health may have been more likely to participate. The employment of a participation incentive sought to reduce this potential bias.
participants were missing from the questionnaire. These foods were various and straddled a range of food groups, although many were vegetarian and vegan in nature. A vegetarian dietary pattern has been reported among other populations (22,24,73) and the vegetarian food items that were reported as missing by more than one participant (tofu; hummus) were therefore integrated into the questionnaire. In addition, due to the possible gaps in food intake identified in the FFQ, an open question was inserted at the end of the survey to ensure total diet was captured in the main study: participants were asked to provide details of any other foods that were eaten on average more than three times per week during the most recent university semester. This enabled identification of participants in the main survey for whom the FFQ was unable to sufficiently capture average food intake, and thus who required potential exclusion from the study (see section 3.5.5).
During focus groups, issues encountered by students whilst completing the web-survey were discussed and ways in which the survey could be improved - to both enhance response rate and data validity - were addressed. Briefly, there was some confusion over the meaning of the FFQ response options and need to complete the questionnaire in terms of average food intake. A study webpage was therefore created detailing the study instructions, which could be referred to at any
time during completion of the survey
(https://sites.google.com/a/sheffield.ac.uk/student-diets/). Concerns were also raised about the applicability of the FFQ to international students: non-EU students commented that many of the foods comprising a substantial proportion of their habitual dietary intake were not included in the survey. Although the FFQ employed in this study (DIETQ, Tinuviel Software Ltd, Warrington, UK) has been relatively validated against 7-day weighed diet records among British adults (228,234), whether it remains valid for non-British individuals temporarily residing in the UK is unknown. It is possible that the dietary data provided by international students might not provide an accurate representation of their habitual dietary intake, thus compromising data validity. It was therefore decided that only Home/EU students would be recruited for the main study. Furthermore, the current project sets out to explore how university life impacts upon dietary behaviour, and inclusion of international students who also have to embrace a new
food culture upon arrival at university would therefore confound this focus. Notably, cultural differences in food intake did not appear to represent an issue for EU students when completing the questionnaire.
Finally, PCA of the dietary data from this pilot study revealed seven major dietary components within the dataset. Importantly, this provided evidence to suggest that university students are not a homogenous group in terms of dietary intake and eating habits, justifying a patterns approach in the main study.