In our children and healthy eating review, we originally planned to extract and synthesise study findings according to our review questions regarding the barriers to, and facilitators of, healthy eating amongst children. It soon became apparent, however, that few study findings addressed these questions directly and it appeared that we were in danger of ending up with an empty synthesis. We were also concerned about imposing the a priori framework implied by our review questions onto study findings without allowing for the possibility that a different or modified framework may be a better fit. We therefore temporarily put our review questions to one side and started from the study findings themselves to conduct an thematic analysis.
There were eight relevant qualitative studies examining children's views of healthy eating. We entered the verbatim findings of these studies into our database. Three reviewers then independently coded each line of text according to its meaning and content. Figure1 illustrates this line-by-line coding using our specialist reviewing software, EPPI-Reviewer, which includes a component designed to support thematic synthesis. The text which was taken from the report of the primary study is on the left and codes were created inductively to capture the meaning and content of each sentence. Codes could be structured, either in a tree form (as shown in the figure) or as 'free' codes – without a hierarchical structure.
The use of line-by-line coding enabled us to undertake what has been described as one of the key tasks in the synthesis of qualitative research: the translation of concepts from one study to another [32 (link),55 ]. However, this process may not be regarded as a simple one of translation. As we coded each new study we added to our 'bank' of codes and developed new ones when necessary. As well as translating concepts between studies, we had already begun the process of synthesis (For another account of this process, see Doyle [[39 (link)], p331]). Every sentence had at least one code applied, and most were categorised using several codes (e.g. 'children prefer fruit to vegetables' or 'why eat healthily?'). Before completing this stage of the synthesis, we also examined all the text which had a given code applied to check consistency of interpretation and to see whether additional levels of coding were needed. (In grounded theory this is termed 'axial' coding; see Fisher [55 ] for further discussion of the application of axial coding in research synthesis.) This process created a total of 36 initial codes. For example, some of the text we coded as "bad food = nice, good food = awful" from one study [56 (link)] were:
'All the things that are bad for you are nice and all the things that are good for you are awful.' (Boys, year 6) [[56 (link)], p74]
'All adverts for healthy stuff go on about healthy things. The adverts for unhealthy things tell you how nice they taste.' [[56 (link)], p75]
Some children reported throwing away foods they knew had been put in because they were 'good for you' and only ate the crisps and chocolate. [[56 (link)], p75]
Reviewers looked for similarities and differences between the codes in order to start grouping them into a hierarchical tree structure. New codes were created to capture the meaning of groups of initial codes. This process resulted in a tree structure with several layers to organize a total of 12 descriptive themes (Figure2 ). For example, the first layer divided the 12 themes into whether they were concerned with children's understandings of healthy eating or influences on children's food choice. The above example, about children's preferences for food, was placed in both areas, since the findings related both to children's reactions to the foods they were given, and to how they behaved when given the choice over what foods they might eat. A draft summary of the findings across the studies organized by the 12 descriptive themes was then written by one of the review authors. Two other review authors commented on this draft and a final version was agreed.
There were eight relevant qualitative studies examining children's views of healthy eating. We entered the verbatim findings of these studies into our database. Three reviewers then independently coded each line of text according to its meaning and content. Figure
The use of line-by-line coding enabled us to undertake what has been described as one of the key tasks in the synthesis of qualitative research: the translation of concepts from one study to another [32 (link),55 ]. However, this process may not be regarded as a simple one of translation. As we coded each new study we added to our 'bank' of codes and developed new ones when necessary. As well as translating concepts between studies, we had already begun the process of synthesis (For another account of this process, see Doyle [[39 (link)], p331]). Every sentence had at least one code applied, and most were categorised using several codes (e.g. 'children prefer fruit to vegetables' or 'why eat healthily?'). Before completing this stage of the synthesis, we also examined all the text which had a given code applied to check consistency of interpretation and to see whether additional levels of coding were needed. (In grounded theory this is termed 'axial' coding; see Fisher [55 ] for further discussion of the application of axial coding in research synthesis.) This process created a total of 36 initial codes. For example, some of the text we coded as "bad food = nice, good food = awful" from one study [56 (link)] were:
'All the things that are bad for you are nice and all the things that are good for you are awful.' (Boys, year 6) [[56 (link)], p74]
'All adverts for healthy stuff go on about healthy things. The adverts for unhealthy things tell you how nice they taste.' [[56 (link)], p75]
Some children reported throwing away foods they knew had been put in because they were 'good for you' and only ate the crisps and chocolate. [[56 (link)], p75]
Reviewers looked for similarities and differences between the codes in order to start grouping them into a hierarchical tree structure. New codes were created to capture the meaning of groups of initial codes. This process resulted in a tree structure with several layers to organize a total of 12 descriptive themes (Figure
Full text: Click here