METHODS

The study used a sequential explanatory mixed methods approach (Creswell & Creswell, 2017) to assess levels of perceived social support from friends among women in southeastern rural Kenya and to gain an understanding of the ways in which participation in an adult women’s soccer league may enhance social support, reduce stress, and improve well-being in this setting. An extensive literature supports the use of mixed methods in instances in which neither quantitative nor qualitative methods alone may adequately capture the trends and details of particular phenomena (e.g., Ivankova et al., 2006; Johnson et al., 2007). The sequential explanatory mixed methods approach used in this study was quantitative-dominant, enabling the study team to use quantitative methods to assess levels of social support among community members and identify significant correlates before specifying the sample and the research questions to be used in the follow-on qualitative phase (see Figure 1). Sport-for-development and social cognitive theories relating to social support provided the conceptual framework for the development of questions in both phases of the study and in the interpretation of findings. Final survey and focus group script content was further guided by the prior field experience of the research team, which included Kenyan nationals with research experience in this setting, researchers with expertise in qualitative methods, and individuals with direct involvement in the formation of the soccer league. The study was reviewed and approved by the Institutional Review Board at Rutgers University and by the National Commission for Science, Technology, and Innovation (NACOSTI) in Nairobi, Kenya.

Figure 1 – The sequential explanatory mixed methods approach

Setting

The study took place in 2018-2019 in Lunga Lunga, a sub-county of Kwale County in the Coast Region of southeastern Kenya. Located on the border with Tanzania, the sub-county is semi-arid, susceptible to extreme drought and flooding, and characterized by limited formal systems of transportation, communication, and financial markets. There are high rates of food insecurity, poverty, and unemployment in the population; many people survive on subsistence farming and herding, income generated through work in the informal sector, and remits from labor migration to more urban areas (County Government of Kwale, 2018). A Kenyan NGO, The Nikumbuke Project, was established in Lunga Lunga as a development initiative to enhance women’s social capital and promote gender equality. Supported by the Swedish nonprofit, From One to Another, and the US-based Health by Motorbike, the Nikumbuke Project provides basic literacy training and health education for its adult female members and school-fee subsidies for their daughters. In addition to these core activities, the NGO offers a training program in tailoring for women who did not complete secondary school and workshops on family planning, gender-based violence, and sexual and reproductive health; the organization has encouraged the construction of town halls in each community as sites for regular meetings of its Nikumbuke members and other local activities, and has worked with communities to install water tanks in areas where access to water is limited. At the time of this study, the Nikumbuke Project was operational in 10 communities that represent a diversity of religions (Catholic, Muslim, traditional, and other) and tribal groups living in seven settled villages, a Maasai pastoralist community, a mixed pastoralist/settled village, and Lunga Lunga, a market town and the seat of the sub-county in which these communities are located. Tribal and ethnic identities are important in Kenya, and communities in this study are generally organized around these attributes. The communities are spread across the sub-county and several are located as far away as 40 kilometers from the Nikumbuke headquarters. Although several villages are situated along the paved roadway used for transporting goods between Kenya’s major port city, Mombasa, and Dar es Salaam in Tanzania to the south, most roads within and between villages are unimproved paths. Residents commonly travel between villages on foot or by motorbike, and paths can become impassable during the rainy seasons for all but foot traffic.

In 2012, an American varsity athlete who was part of a six-week university study abroad program affiliated with the Nikumbuke Project decided to play pick-up soccer with the local children as part of her exercise regimen. To everyone’s surprise, the women members of the Project, some of whom had either played soccer as students themselves or had children currently playing soccer in school, enthusiastically joined in. Later that year, the Nikumbuke Project membership requested that it be given the opportunity to start its own soccer league. Believing that such a league would contribute to its health and wellness goals, the Project leadership asked the same American student for assistance in setting up the new league. In 2014, she established a US-based nonprofit, the Nikumbuke Soccer League, to offer guidance and support for an all-women’s adult recreational soccer league, the first of its kind in Kenya. In 2019, at the time of the study, each of the ten villages in the Nikumbuke Project hosted a team of 22-23 players selected by the Nikumbuke members in that village.

Quantitative Phase

A semi-structured survey instrument was used to collect cross-sectional data in the quantitative phase of the study. It contained questions relating to demographics, health, gender norms, soccer participation, and access to social support. Before the survey launch, members of the research team attended a regular meeting of each village group to introduce the study and develop a contact list for follow-up recruitment. Prior to the start of data collection, project members in each community were asked to nominate individuals to serve as a member of the local data collection team. From these nominees, the study team assembled a team of women who were able to read and write and were fluent in Swahili; many of them had experience serving as “teachers” in the Project’s regular literacy programs. They participated in a week-long training program in research ethics and survey techniques run by the study team and practiced both consent procedures and survey administration using role-play exercises. Data collectors received a small stipend for successfully completing their training (500 Kenyan shillings [KES]—approximately $5 in U.S. dollars) and for each survey they administered (300 KES). The contact list created at the introductory meetings in each community was used by the study team to schedule individual visits with women who were interested in participating. Women who agreed to participate in the survey were also given a small compensation (200 KES). The surveys took between 35-65 minutes to complete and were administered by the local data collectors in a private setting, quiet corners of the village halls, or respondents’ homes. Surveys were checked by members of the study team at the end of each day to flag any methodological errors and to identify any data collectors who might be having difficulty with aspects of survey administration. Although available in both English and Kiswahili (the official languages of Kenya), the majority of surveys were administered in Kiswahili. Data collectors in the two communities that were home to Maasai were also able to translate questions verbally into Maa (the language of the Maasai people) as needed. Two members of the study team who live in Kenya and are fluent in Kiswahili and English, moved among the ten communities at the beginning of and throughout data collection and were available to answer questions, review completed surveys, and provide guidance where needed.

Sample

All 702 women who were members of the Nikumbuke Project, including members who played soccer in the league (229) as well as those who did not (473), were invited to participate in the quantitative portion of the study, and there were no refusals recorded. To be eligible, members had to be 18 years or older and able to provide verbal informed consent. The local data collectors, who had completed the surveys as part of their training, were excluded from the final sample of 684 women.

Measures

The outcome variable of interest was perceived social support from friends (PSF), measured by women’s responses to five questions about their perceived ability to access different kinds of support from friends when needed: emotional support, food, money, shelter, and someone with whom to talk about important matters. Responses to each item were recorded on a three-level Likert scale reflecting agree, partially agree, or disagree. Women were asked whether they participated in other groups besides the Nikumbuke Project, and to indicate their level of agreement with two additional questions: “most of my friends come from my village/community”, and “I have friends in other villages/communities”. Women were also asked to indicate, using a five-point scale ranging from none of the time to all of the time, the frequency in the past four weeks in which emotional or physical health had interfered with social activities. Sociodemographic measures were also included to record age, number of children, relationship status, education, village of residence, and household headship. Based on reported village of residence, a variable type of village was created to indicate whether the respondent resided in the pastoralist community, a settled village, the mixed pastoralist/settled community, or the market town.

Analysis Plan

Data analyses were conducted using StataCorp statistical software, Release 16 (StataCorp, 2019). An exploratory factor analysis (EFA) was conducted to determine if the five separate questions relating to whether a woman had friends on whom she could rely for various resources could be used as a single measure of perceived support from friends (PSF). For this purpose, listwise deletion was used to eliminate those respondents for whom responses to one or more of the five support questions were missing; this resulted in an EFA-analytic sample of 673 respondents. The sample was found to meet the threshold for factor analysis as recommended by Hair et al. (1998), with an average inter-item correlation of .37, Kaiser-Meyer Olkin Measure of Sampling Adequacy of 0.75, and significant Bartlett’s Test of Sphericity (χ2 (10, n=673) = 744.23, p < .001). Principle components analysis revealed only one factor with an eigenvalue greater than one, suggesting a simple structure with all five items loading strongly onto that factor. This factor explained 50% of the variance with factor loadings from .65 to .79. The five-item scale had a Cronbach’s alpha of 0.75. (see Table 1).

Table 1 – Items included in ‘perceived support from friends’ (PSF) scale*

* Possible values of responses for individual items ranges from 0 to 2, with 2=Agree, 1=Partially agree, and 0 = Disagree

Each item had a value ranging from 0 to 2; the scale was scored as a summed total of the values for all five items. For bivariate and multivariate analyses, a binary variable perceived social support was created to reflect whether respondent scores on the PSF scale were lower than/equal to or higher than the mean score of the sample. Frequencies were computed for all study variables. Separate Pearson’s chi-square tests of independence were run to explore bivariate relationships between level of perceived social support, participation in the soccer league, and sociodemographic attributes.

Two multivariate regression models were used to examine the relationship between women’s PSF and their participation in soccer. The first was a linear model containing the continuous variable PSF, with values ranging from 1 to 15 as the dependent variable; the second used a dichotomous variable indicating social support that fell either above the mean/at or below the mean. An item non-response analysis of the sample surveys indicated minimal missing data on all variables less than 5% except for age (9.5%). Missingness among the five items comprising the PSF scale ranged from 0.29% to 1.2%. For the regression analyses, multiple imputation was used to approximate missing data for all variables included in the model. Values for age were not imputed given the large percentage of missing values (9.5%), nor was age, given its lack of statistical significance at the bivariate level with perceived social support, included in the multivariate models. Variance inflation factors (VIF) for all variables included in the models were less than 1.5, well below the generally accepted threshold indicating multicollinearity (Hair et al., 1998). The strength and direction of the relationships between the independent and the dependent variables were similar in both the linear and logistic regression models. As many of the independent variables were categorical, the logistic regression model was somewhat easier to interpret and those findings are reported here.

Qualitative Phase

The qualitative phase took place five months after collection and preliminary analysis of the quantitative data. Exploratory sequential mixed methods research utilizes findings from an initial quantitative phase to shape the design, sample, and questions that will comprise a follow-on qualitative phase (Creswell & Creswell, 2017). In this study, the quantitative data suggested a significant relationship between women’s participation in the soccer league and high social support (see Table 2). To explore this relationship further, the qualitative phase of the study involved a purposeful sample that included only women from the Nikumbuke Project who played in the soccer league.

All members of the soccer league (N = 229) were invited to participate in the focus group discussions (FGDs), which were facilitated in Kiswahili by the third and fourth authors. At the start of each FGD, the facilitators explained the purpose of the FGD, how the session would be organized, and asked for women’s oral consent to participate and be video recorded for the purpose of transcription. Each participant received a modest compensation of 100 KES for their time. Although the FGDs were to be facilitated primarily by the study team, it was decided that Nikumbuke members who had served as data collectors during the quantitative phase of the study would benefit from the opportunity to receive training, assist in the coordination of the discussions and, where needed, translate into the Maasai language. Training sessions were held to review study goals and objectives, ethical issues relating to focus groups, and qualitative data collection. Trainees received a stipend of 300 KES for successful completion of the qualitative training program and for their participation in focus-group facilitation.

Two focus-group discussions of 10-13 players each took place in each of the ten Nikumbuke communities to allow all members in that community’s soccer team the opportunity to participate without the FDGs becoming unwieldy. Each FGD lasted approximately 90 minutes. Sessions followed a script containing open-ended questions about women’s motivations in joining the soccer league and the ways in which playing on the team had affected their general health, their relationships, and their roles in the community. Eighteen of the sessions were conducted in Kiswahili and two in the Maasai language.

Video recordings of sessions conducted in Kiswahili were first transcribed into Kiswahili and subsequently translated into English by a transcription service in Nairobi. Video recordings of the two sessions conducted in Maa were simultaneously transcribed and translated directly into English by a language instructor in Nairobi fluent in Maa, Kiswahili, and English. An error in videotaping during two of the focus groups necessitated the omission of these sessions from the analysis. The resultant analytic sample included data from two FGDs comprising members from each of eight teams and one FGD from each of the two teams in which the videotaping errors had occurred, for a total of 201 women. All qualitative analyses were performed in NVivo 12 using the English translations (QSR International, 2018).

Analysis Plan

The qualitative phase used a thematic analysis approach to identify and interpret patterns in the data (Braun & Clarke, 2006). The study team members responsible for analyzing the qualitative data included the first author (the project’s primary investigator, who has over seven years’ experience in working with the Nikumbuke Soccer League), and the second author, a team member familiar with the setting and an experienced qualitative researcher. These authors, neither of whom speak Kiswahili, consulted with team members who were fluent in Kiswahili in those instances where interpretation of the underlying meaning of a translated phrase was not clear. First, they familiarized themselves with the data by reviewing each transcript, drafted a list of initial codes independently, and then met to generate a common set of codes. Next, each reader coded two focus groups independently and then compared coding choices to review and refine the codebook. Using this revised codebook, each reader independently coded six of the focus groups to assess interrater reliability. Interrater reliability (Cohen’s kappa) was, on average, 97% and ranged from 94-99%. The two readers met to align code definitions. Both then independently finished coding the remaining focus groups, discussing issues as they did so. Focusing on the theme of social support, the two met frequently to further refine, name, and define the theme and sub-themes, and to ensure interrater reliability. Qualitative findings were then presented to the remaining five authors, including those authors who had facilitated the focus groups (third and fourth authors). All authors collaborated on the interpretation of the findings.