This study aimed to analyse whether increasing the number of key informants from 4–6 to 12–15 alters the results of the CRA. Overall, there were no large differences in the CR scores between the initial and the extended sample. Extending the sample did not change the mapping of the communities to a certain stage of change. The CR scores showed little variation after having interviewed 10 key informants in a community, indicating that this may be a good number to strive for in CRA interviews. Our results are similar to a study by Guest et al. [10] who systematically examined the degree of data saturation in 60 in-depth interviews with women in two African countries. The authors started with analysing the first six interviews, then added a set of six more interviews, and so on. The process was repeated until all 60 interviews have been analysed. The authors found that data saturation occurred within the first 12 interviews with 92% of codes identified therein [10]. Two approaches on data saturation, code saturation (i.e., no additional issues raised, codebook stabilized) and meaning saturation (i.e., no further nuances or dimensions can be found), were examined in 25 in-depth interviews with HIV patients [9]. The authors found that code saturation was reached after nine interviews (91% of codes identified) and meaning saturation was achieved between 16 and 24 interviews. In addition, a study by Francis et al. [8] found that data saturation (i.e., minimum of 10 interviews, no new themes after analysing 3 further interviews) in theory-based interview studies was achieved after conducting 17 interviews. Focusing on studies applying the CR tool, Kostadinov et al. [6] included 40 articles published between 1999 and 2013 in a systematic review. The authors reported that the number of interviews per community varied between one and 33 (mean = 7.3, SD = 5.9, median = 6, six studies provided no information on the number of interviews per community), with 15% of studies using less than the recommended four interviews per community. This is in line with studies published since 2013 that followed the recommendations of the CR handbook and conducted an average of four to eight interviews per community (e.g., in [15–19]). However, Kesten et al. [20] used the concept of theoretical saturation (i.e., no new concepts are expected by conducting more interviews) to determine the necessary number of interviews for assessing CR for overweight and obesity prevention in pre-adolescent girls. Theoretical saturation was reached after conducting 33 interviews with key informants (e.g., parents, teachers). In addition, Pradeilles et al. [21] interviewed 12 key informants per community across two cities in Ghana, but did not give any reasons for selecting 12 interviewees per community. While it seems that conducting 10 interviews per community provides a fairly stable classification of a community in terms of their CR score, our results also indicate that there may still be unidentified issues regarding community efforts for PA promotion, contact and communication channel for informing or approaching the target group, or barriers to participation in PA promotion. A reason for this could be the nature of the PA efforts in the communities. There were many small scattered efforts and thus it appeared that the key informants did not mention all or were not aware of them. This indicates the context- and topic-relatedness of CRA, which needs to be kept in mind when planning similar studies.
Strengths and limitations
The main strength of our study is that we conducted more than the recommended 4–6 interviews per community to obtain a full picture of the CR for PA promotion in older adults. However, our study has some limitations. Assuming stability of a mean if the change is less than 0.1 SD with every new respondent is an arbitrary choice. One could argue that small differences can lead to large changes if the sample is large and the differences go into the same direction. However, the results in our four communities did not indicate such a monotone pattern concerning the direction of change. Furthermore, there is a natural limit to the sample size of key informants. The number of potential key informants depends on the topic, the existing structures in the community, and the size of the community. For instance, in one of our selected communities it was not possible to identify more than 12 key informants who could provide information on community-based PA promotion in older adults. Another limitation is that in some cases key informants answered from their own point of view and may have had difficulties to respond on behalf of their community.