Development and Validation of the Soweto Stress Scale: Mixed-Methods, Population-Based Study of Adults Living in Soweto, South Africa

Abstract

Aim.

This study uses mixed methods to generate a locally specific assessment of stress and evaluates how it relates to general psychiatric morbidity.

Methods.

We conducted two related ethnographic studies of stress and distress in Soweto, South Africa. We then used these studies to develop the Soweto Stress Scale, piloted the scale, and evaluated it in an epidemiological surveillance study of stress and health. We used factor analyses to evaluate item relatability within the measure and maximum likelihood estimates to evaluate comparative fit indices.

Results.

The Keiser-Meyer-Olkin test identified sufficient sample variation and the scale was suitable for factor analysis. The confirmatory factor analysis of the 10 items supported a single-factor model with a χ2(119) =527.78; p< 0.01. Even though the comparative fit index was relatively poor and could not be improved, internal consistency (Cronbach’s α=0.81) was excellent, suggesting that the scale is reasonable. There also was covariance between the scale and stress measured with other scales.

Conclusions.

The Soweto Stress Scale reports strong internal consistency and reflects a very localized view on social stressors among individuals and may serve to identify those with higher psychological morbidity. Given the racial, ethnic, and linguistic diversity within South Africa, our “emic” stress scale was tested in a community sample but may be useful for screening individuals with higher levels of stress or psychological distress within clinical settings.

Introduction

More than a decade ago anthropologists began calling for thoughtful ways to translate “key insights from ethnography into tools that can be used by epidemiologists, demographers, and other population health researchers concerned with causal inference regarding human behavior, biology, and health” (Brown et al., 2009: 248). Doing so requires creative ways to think through the impact of psychosocial phenomena identified through ethnographic work to contribute to existing epidemiological methods that can provide a more nuanced understanding of local context and experience. To achieve such contributions, anthropologists must work together with epidemiologists to communicate how people define certain experiences (while keeping culture in mind) and to quantify them into markers of experience. In turn, such work can produce new questions for ethnographic research.

Moving from ethnography to epidemiology can be tricky. First, as most epidemiological and demographic studies do, researchers either utilize internationally or locally validated measures to assess what are usually very complex psychosocial constructs or, less commonly, create their own surveys if existing scales are inadequate, although efforts to adapt scales are not always effective (Weaver & Kaiser, 2015). Second, as anthropologists have previously argued, numbers and global health metrics can flatten the complexity of lived experience and social and political realities (Adams, 2016; Sangaramoorthy & Benton, 2012). Meaningful histories of segregation, stigma, and political violence can be erased when research assistants check off a series of Likert scale responses (e.g. 1–5, not concerned to very concerned), and in the context of biological studies, these histories can be reduced to concentrations of stress hormone levels, body-mass index, and cognitive function, opening the door for victim-blaming narratives (Authors).

Third, the subjectivities of the researcher can also bias methods of data collection, statistical modeling, and what is seen as “traumatic,” causing a privileging of the researcher’s worldview over those of the participants. Stressors and traumas that become biologically and socially embodied and intergenerationally transmitted must first be appraised as stressful or traumatic, which is inherently a culturally-mediated process (Kohrt et al., 2009). Thus, poor assessment of psychosocial experiences like “stress” can obfuscate deep political dimensions, present missed opportunities for observing biological responses to trauma, and ultimately undermining the search for empirical “truth.”

Yet, it is rarely harmful to obtain better measurements of complex psychosocial phenomena, and for epidemiologists and public health researchers, this undertaking requires engaging in deeper ethnographic theory and practice and producing reflexive scientists. Deeper ethnographic research on trauma and mental health, for example, can lead to the development of surveys that are more sensitive to the cultural realities of larger communities. For example, Kaiser et al. (2013) utilized a rigorous ethnographic approach consisting of long-term participant-observation, in-depth interviews, and focus groups to adapt existing depression and anxiety screeners and develop new mental health measures that better accounted for locally salient symptoms of distress and their negative sequelae in the Central Plateau of Haiti. Certainly there are other examples where researchers designed locally defined surveys instruments to capture common ways of conveying stress and distress to identify those with undue distress and duress (Ashaba et al., 2019; Knettel, 2019; Kohrt et al., 2009; Rasmussen et al., 2015; Van Ommeren, 2003; Van Ommeren et al., 1999). Indeed, such approaches have been common in public health research for decades (Campbell et al., 2000; Onwuegbuzie, Bustamante, & Nelson, 2010; Waldram, 2006).

Assessing Stress in Soweto

This study was conducted in Soweto, a community situated in the southwest corner of Johannesburg, South Africa, that is characterized by a wealth of ethnic diversity and integrated linguistic and cultural nuance reflected in the families who have resided there for generations. Soweto has its own unique cultural milieu, with families often speaking one or more languages, from IsiZulu, IsiXhosa, Setswana, Sesotho, and Xitsonga to English and Afrikaans, and is relative wealthy compared with other “townships” – reflecting its history as an incorporation of six racially segregated urban areas amidst apartheid and the transformative decades that followed.

Today Soweto is a flourishing metropolis. Yet, social and economic inequality have produced an undue burden of multiple conditions – a quadruple burden of disease – within Soweto and other low-income and racially segregated neighborhoods (Coovadia, Jewkes, Barron, Sanders, & McIntyre, 2009; Pillay-van Wyk et al., 2016). Recent estimates show that the prevalence, incidence, and burden of mental illness in South Africa are relatively high compared to other countries worldwide: about one in three South Africans are diagnosed with a mental illness, over a quarter of which are considered severe, and nearly half of all citizens will develop a psychiatric disorder in their lifetime (Herman et al., 2009). And in many cases, such psychiatric conditions are not only associated with chronic physical illness – such as HIV and Type 2 diabetes – but also with social stressors such as food insecurity and interpersonal violence (Abayomi & Cowan, 2014; Dewing, Tomlinson, Roux, Chopra, & Tsai, 2013; Karim, Churchyard, Karim, & Lawn, 2009; Mayosi et al., 2009; Oni et al., 2015; Tsai, Tomlinson, Comulada, & Rotheram-Borus, 2016b, 2016a). Thus, the extraordinary psychological burden from epidemic stress (HIV) as well as political stress (apartheid), have played a fundamental role in how people identify and experience stress, and provides an exemplar context to study how what people think about as stressful in their daily lives can impact mental illness.

This article investigates the development and utilization of an ethnographically-generated stress scale, the Soweto Stress Scale. The studies provided emic, or non-biomedical cultural frameworks, for understanding what and how life stressors caused undue burden to people’s physical and mental health (as opposed to a pre-determined or internationally standardized checklist of what people may find to be stressful, which may be defined as etic). Hinton and colleagues (2016) have argued that how people experience and express stress or distress convey varied social, cultural, political, and somatic factors insofar as they go through a process of “localization” (which Hinton uses specifically to discuss “thinking too much”—a common idiom of distress). Thinking about how certain types of social stress and psychological distress may be localized in Soweto, or similar regions elsewhere in South Africa, is imperative when conveying what may fuel mental illness and/or impede people from flourishing.

In what follows, we describe the methods we used to carefully devise the Soweto Stress Scale. We introduce two separate but related ethnographic studies from which we drew to develop the scale and what methods we used to then evaluate the psychometric properties of the scale in a population sample. Moving from ethnography to epidemiology, in this case, is imperative not only to better understand the broad relevance of ethnographic evidence but also to inform what factors drive what is deemed as “good mental health” or “poor mental health” in population studies, and then to reflect back upon what we learn from people’s lived experiences.

Methods

Study setting and sampling

This study was nested within the Developmental Pathways for Health Research Unit, a research unit under the South African Medical Research Council (SAMRC) and the University of the Witwatersrand located at Chris Hani Baragwanath (“Bara”) Academic Hospital in Soweto. The research site served as home base for the research team who conducted interviews in people’s homes. All research participants were residents of Soweto. The University of the X Institutional Review Board approved the three studies discussed in this article in three unassociated, distinct applications (2012, 2017, 2018) and each study participant engaged in informed consent before engaging in the studies. Although the first two studies were conducted at the research station at Chris Hani Baragwanath Hospital, the larger surveillance study was conducted in people’s homes throughout Soweto proper.

Step 1: Developing the Soweto Stress Scale

The Soweto Stress Scale was developed from 107 in-depth, life history narrative interviews conducted with people residing in Soweto. These interviews were conducted in two separate studies, with the first study in 2012 focused on stress and diabetes and the second study in 2017 focused on stress, cancer, and comorbidity (Authors). These studies involved lengthy life history narrative interviews, each lasting between two and four hours long, which focused on myriad factors throughout study participants’ lives: where they were born, early life, family, school, early adulthood, marriage, childbearing and childrearing, paid and unpaid work, communities and neighborhoods, migration, religious life, home environment, and family. Most conversations focused on family dynamics and personal experiences within their home, work, neighborhood, and church. We probed about idioms used to describe stress and distress, perceptions of living with chronic illness, and experiences facing life with multiple cascading medical and social conditions. All participants were adults between the ages of 30 and 75 years, and most were women (77 [72%]).

We conducted, transcribed and translated, coded, and analyzed each 2–3 hour interview and analyzed the narratively-emergent themes described by study participants (see Authors). Separately as two independent studies and then together as a collective of 107 life stories, we listed every stressor that was reported and coded from each study. We identified the emergent codes about stress (from the original code books) and revisited how the codes were defined in each code book. Then we reviewed specific texts that were coded using those codes to ensure that we agreed that each code was in fact distinct from others and meaningful in both datasets. Then, we examined how frequently these codes emerged across the datasets. We also evaluated (as we have in related published work) how many people were likely to report symptomatology of depression and report the stressor. Because these were small studies, we did not have any conclusive data of causation or correlation.

The 21 most common stressors, which emerged in both studies, were included in the Soweto Stress Scale. We show each variable in Table 1 and provide a corresponding quote to reflect broader meaning of the item, as well as a citation where the variable was described in detail. We provide unpublished data for some items, and these can be contextualized more in a related book (Authors). There was largely agreement among the two studies in terms of what types of stressors were common among those we had interviewed; however, since we interviewed 77 women and only 30 men, our knowledge about women’s experiences is greater than our knowledge about men’s experiences.

Table 1.

Socio-demographics of sample (n = 933).

Variables		Mean (SD)	%	Range
Demographics
Gender (% female)	522		67.1
Age (at enrollment)		46.0 (12.7)		26–71
Educational attainment (% attended)
No school or primary school	482		62.3
Secondary school	230		29.7
Professional/teaching/university	62		0.08
Household assets		7.9		3–12
Psychological status
Soweto Stress Scale		46.9 (12.9)		21–92
General Health Questionnaire		54.9 (10.3)		18–90
GHQ Somatic Symptoms		1.65 (1.49)
GHQ Anxiety and Insomnia		1.74 (1.60)
GHQ Social Dysfunction		2.6 (0.83)
GHQ Severe Depression		0.71 (0.05)
GHQ Total		2.94 (3.62)

Step 2: Administering the Soweto Stress Scale

We nested the present study in the enumeration study, where we obtained a simple random sample of geographic coordinates (latitude and longitude) within residential areas of Soweto, which we identified as “clusters”. There were six clusters in total. For each of the randomly selected coordinates, the closest dwelling within 30 meters was approached for enrollment and identified as part of that cluster. Of 2,000 coordinates visited in Soweto, 11% did not have a dwelling within 30 meters. Of the thousands of households approached, 86% consented for participation in the research study.

For our study component we had two field teams (with two research assistants each) conducting interviews. Inclusion criteria included: people 25 years of age or older who lived within each identified cluster and who considered themselves to be a resident of Soweto. We interviewed those 25 years of age and older so that our recruitment would not impede another concurrent study that was focused on interviewing young adults. All people younger than 25 years of age were excluded as well as individuals who could not meaningfully communicate with the study team, such as those with cognitive impairments, were intoxicated, people who were too ill, or people who threatened our team.

Our field teams obtained written informed consent within people’s homes and then conducted the survey interviews on tablets, where they imputed survey responses directly into Research Electronic Data Capture (REDCap). These interviews involved questions on socio-demographics, stress and coping, adverse childhood experiences, social coping and cohesion, medical histories, and anthropometrics. For the Soweto Stress Scale, we asked study participants how often they felt or experienced stress associated with certain experiences in the four weeks prior to the survey. The possible response options were: 1) you have never felt the stress in the past four weeks, 2) you have seldom felt the stress; 3) you have often felt or experienced the stress; 4) you have very often felt or experienced the stress; and 5) you always feel or experience the stress. We interviewed 957 participants, of whom 791 had full data on all variables of interest and were therefore included in this analysis.

Step 3. Evaluating the psychometric properties of the Soweto Stress Scale

The total sample was split randomly into two halves to explore and analyze the construct structure and validity of the Soweto Stress Scale. We used the first half of the sample (n=467) for exploratory factor analyses (EFA). We used the Keiser-Meyer-Olkin (KMO) test to check for sampling adequacy. KMO values between 0.8 and 1 indicate sampling adequacy, values < 0.6 indicate that the sample is inadequate, while KMO values close to zero indicate a widespread correlation. As part of the unrotated EFA we included a post-estimation command (estat anti) to check for variables that may correlate too highly with each other. As Field (2013) suggests, inspection of the correlation matrix before proceeding might help identify potentially problematic variables and to ensure that all factors included in the rotated model have unique contributions to a factor (or factors) extracted. We also conducted a parallel analysis to confirm the results shown by the scree plot in the EFA.

Because we were trying to understand the structure of variable clusters or identify latent variables, we used the principal factor (pf) estimation technique. We chose the oblique oblimin rotation, which allows factors to be correlated and produced the simplest solution.

We used the second half of the sample (n=466) to conduct confirmatory factor analysis (CFA), using maximum likelihood (ML) estimation to explore the goodness-of-fit of the exploratory models. Fit indices calculated were: chi-square (χ²), chi-square/degree of freedom ratio (χ²/df), the comparative fit index (CFI; Bentler, 1999, the Tucker-Lewis index (TLI; Bentler, 1999, the root mean square root of approximation (RMSEA; Steiger, 1990), and a standardized root mean square residual (SRMR; Hu Bentler, 1999). Best practice guidelines suggest that χ²/df should be less than 5; SRMR should be close to zero; and RMSEA should be <.05, thus indicating a close fit, whereas <.08 indicates a reasonable model, and values exceeding that indicate a mediocre or a poor fit. For a good fit, the CFI and TLI are recommended to be ≥0.90 (Bentler, 1999; Byrne, 2010).

We conducted a likelihood ratio test (LRT) to compare a 2-factor model (m₁) to a single factor model (m₂). The assumption was that m₂ is nested in m_1, and the purpose was to find a statistical justification for choosing the single-factor model. The LRT compares the log likelihood values of the two models, and if the difference is statistically significant then the less restrictive model (m₂) is said to fit the data better.

We assessed internal consistency and reliability of the extracted factor (i.e., CFA results) by using Cronbach’s α. Overall mean scores for each measure were tabulated. Pearson’s correlations were used to estimate the associations between the Soweto Stress Scale and the 28-item General Health Questionnaire (GHQ-28). Stata 15 was used for analyses (StataCorp LLC, College Station, Tex.).

Results

Table 1 shows that the mean age was 46 years (range, 26–70 years). Most participants (579 [62.0%]) completed primary school and 251 (27%) completed secondary school. The asset checklist revealed socioeconomic diversity: On average people reported 8 items or more, but variation was notable (range, 3–12 items). The overall mean score for the Soweto Stress Scale was 26.4 (standard deviation [SD], 13.0; range, 0–71). The items derived for the Soweto Stress Scale are described, with sources notated, in Table 2. The means and standard deviations of the items of the Soweto stress scale are presented in Table 3.

Table 2.

Deriving scale constructs through qualitative research

Item from the Soweto Stress Scale	Representative Quote and Source
How often do you feel stress associated by…
Having enough money to cover basic needs?	“I only make a few cents from the beadwork and I supplement that with the social grant that I receive from the government.”
The cost of food?	“I do not have the money to buy the necessary food that I should be eating. The right food is very costly.”
Having enough food to eat?	“I eat one fruit per day, and I should eat four, but I can only afford one.”
Your children’s futures?	“What stresses me the most is that I have been raising orphans. There is K who accompanied me here and his siblings, they lost their mother. I think God because these kids never gave me problems, I used to get stressed when I looked at them and times were tough but they would also comfort me.”
Your family member’s safety?	“My nephew bullies and abuses everyone in the household, especially when he is drunk.”
Walking alone at night?	“They [tsotsis] have burgled my house on two occasions and I have once been mugged while on the road and I lived in fear ever since.”
Losing a family or friend to violence? (e.g. guns/clubs?)	“He [my son] was gunned down on his way from work. He was a police officer and he tried to stop a fight only to be shot at by the same people.”
A spouse or partner who takes too much alcohol or drugs?	“I can say that it was my husband. We fought a lot and he had a drinking problem. We would fight over children and every other thing.”
Thinking about a past abuse from a family member?	“I was never appreciated and at times he [my husband] would hit me for my cooking. He would also accuse me of promiscuity. He would at times refuse to look after his kids and claim that there weren’t his. That on its own was abusive.”
Feeling unsafe around a family member?	“I was never appreciated and at times he [my husband] would hit me for my cooking. He would also accuse me of promiscuity. He would at times refuse to look after his kids and claim that there weren’t his. That on its own was abusive.”
Needing to seek medical care?	“The clinic is ill-equipped because at times there aren’t enough gadgets to test either our sugar levels or just our blood pressure. At times the toilets are out of order.”
Following the doctor’s orders for self-care?	“I like meat with all my heart, I must be honest. So, it’s even sometimes difficult for me to remove the fat; it becomes so tasteful I must be honest, it’s very difficult for me to leave fat [… and] to eat plain food without fat. Again, I had a problem leaving our stable food “pap” as prescribed by the doctor.”
Caring for a family member who has an illness?	“[My son] was sick and died here at Bara. He had heart and kidney trouble; also his lungs were not fine because he drank a lot. […] My daughter died in 2009, and then my husband died in 2010. She [my daughter] had a bad liver. […] My husband then died in 2010 at the age of 76. He had kidney problems, high blood pressure, and diabetes just like me.”
Your job?	“I speak my mind out at home but I suppress my feelings at work in order to accommodate my colleagues. It is easier to reason with a kid than a colleague. Sometimes colleagues can be very difficult and leave you with very little option but to do their work especially when you are management.”
Not having a job?	“I lost my job in 1996 as the company was closing down and I was 40 years old at the time. Maybe I couldn’t get a job anywhere else because of my age. I eventually gave up trying to find a job anyway.”
Memory of something that happened to you during your childhood?	“My stepfather sexually abused me. […] He would force himself on me and when I told my mother the man would accuse me of things that I never did. My mother would get upset and punish me.”
Feeling sad?	“I feel tired and sleepy and try to isolate myself.”
Thinking too much?	“I think too much and become so stressed.”
Worried about having an illness?	“I am always sick, all the time I go to the hospital. My mother feels so sad because of my conditions. She has taken my only child away from me [because] she does not want him to see me this way. I feel so bad too because I want to live with my son.”
Feeling stress or shame caused by judgment by others about your illness?	“My siblings are isolating me because of my illnesses Even if I had my line for washing, they don’t want to hang their clothes where I hang mine. Everything they do is to isolate me.”
Feeling physical pain or discomfort?	“I sometimes get stressed and I feel pain around my shoulders and the back of my neck. I hardly get any sleep when I am in that condition and I feel the same pain around the shoulder area. I also get very edgy and easily irritable when I am stressed.”

Table 3.

Exploratory factor analysis (EFA) loadings.

Items F-I	Item (brief content)	KMO	Factor Loading
1	Having enough money to cover basic needs?	0.84	0.57
2	The cost of food?	0.85	0.59
3	Having enough food to eat?	0.81	0.47
4	Your children’s futures?	0.83	0.44
5	Your family member’s safety?	0.81	0.38
6	Walking alone at night?	0.82	0.31
7	Losing a family or friend to violence? (e.g. guns/clubs?)	0.82	0.37
8	A spouse or partner who takes too much alcohol or drugs?	0.79	0.24
9	Thinking about a past abuse from a family member?	0.86	0.46
10	Feeling unsafe around a family member?	0.83	0.38
	Eigenvalue		4.02
	M		1.38
	SD		0.68
	Cronbach’s alpha coefficient		.81
11	Needing to seek medical care?	0.79	0.38
12	Following the doctor’s orders for self-care?	0.48	0.12
13	Caring for a family member who has an illness?	0.92	0.40
14	Your job?	0.48	−0.08
15	Not having a job?	0.72	0.39
16	Memory of something that happened to you during your childhood?	0.81	0.48
17	Feeling sad?	0.83	0.67
18	Thinking too much	0.86	0.71
20	Feeling stress or shame caused by judgment by others about your illness	0.84	0.31
21	Feeling physical pain or discomfort	0.81	0.40

Factor analysis

We checked the correlation coefficients in an attempt to check for variables that may correlate too highly with each other, and there were no variables with correlations >.90. For the EFA, we extracted factors using the principal factor method. The overall KMO was 0.82, and the scree plot suggested a single dominant factor (Figure S1 in online supplement). We also conducted a parallel analysis, and the plot represented by the dashed line crossed the solid line just after the sixth factor-point (Figure S2), suggesting that variables were loading onto six or more factors (Figure S2). The pattern matrix showed that the third through sixth factors were only loading two variables each. After testing 1-, 2-, and 3-factor solutions, we opted for a single-factor solution: Table 3 presents the factor loadings. The items “a spouse or partner who takes too much alcohol or drugs”, “following doctor’s order”, and “having a job” were below the factor loading cutoff of 0.30 and were initially retained. The estimated Cronbach’s alpha for the single-factor solution was α = 0.81.

Confirmatory factor analysis

Table 4 depicts the results for the final single-factor CFA model using the second split-half sample. The single factor solution had poor fit (CFI=0.67, TLI=0.63) and a RMSEA of <.08. To improve the model, we deleted the items “a spouse or partner who takes too much alcohol or drugs”, “following doctor’s order”, “having a job”, and “walking alone at night”. The first three were below the factor loading cutoff in the EFA analysis, and “walking alone at night” was deleted last because it had a loading close to the cutoff (0.30). The resulting model had a slightly improved fit, but the fit indices remained poor (CFI=0.74, TLI=0.72). However, the RMSEA was <.08 and thus showed a reasonable fit. The LRT results showed that the single-factor model (m₂) fit the data better: the two-factor model (m₁) statistics were χ2(103) = 285.12 and p < 0.001, while the single-factor model (m₂) statistics were χ²(104) = 481.33 and p < 0.001. The LRT statistic was χ2(1) = 196.20, p < 0.001. Therefore, the single-factor model was selected.

Table 4.

Goodness of fit for the Soweto Stress Scale.

Goodness of fit indices
Models	χ2(df)	χ2/df	SRMR	RMSEA	CFI	TLI
Soweto Stress Scale (unadjusted)	733.65(170)	4.31	0.07	0.08	0.67	0.63
Soweto Stress Scale	475.88(104)	4.57	0.07	0.08	0.74	0.72

Internal consistency and construct validity of the CFA model

Even though the single factor solution had poor fit indices, the reliability of the Soweto Stress Scale was excellent (α = 0.81), which was consistent with the EFA estimation. To assess construct validity, we used the four domains of the GHQ-28 (Somatic Symptoms, Anxiety, Social Dysfunction, and Severe Depression). As hypothesized, three GHQ-28 domains were positively correlated with the Soweto Stress Scale: somatic symptoms, anxiety, and severe depression (Table 5).

Table 5.

Correlations of the latent variables of the stress checklist measure with subscales in the general health questionnaire (GHQ).

Measure	1	2	3	4	5
1. Soweto Stress Scale	-
2. GHQ Somatic Symptoms	0.44	-
3. GHQ Anxiety	0.52	0.51	-
4. GHQ Social Dysfunction	0.05	0.19	−0.00	−
5. GHQ Severe Depression	0.30	0.41	0.48	0.19	-

Note. Coefficients in boldface reach statistical significance, p<.05.

Discussion

In this mixed-methods, population-based study of adult South African men and women, we used ethnographic methods to create candidate items to measure study participants’ experiences of stress, and then we applied psychometric analyses to data from a population-based survey to create the Soweto Stress Scale. Our analyses indicate that the ethnographically derived measure of social stress has a coherent factor structure and acceptable internal consistency, with strong evidence of construct validity given its correlations with various domains of psychiatric morbidity from a widely utilized assessment of mental health status. Taken together, these findings suggest that the ethnographically designed Soweto Stress Scale may be a useful tool for research on psychosocial wellbeing and mental health.

This analysis writes against a racist history that used White psychology as the norm. During apartheid, psychological tests travelled to South Africa from Britain, and followed a similar pattern to the US, where etic tests were widely applied to a population with diverse histories, languages, and cultures (Claassen, 1997). In some cases, these tests were utilized to determine who would access certain jobs, educational opportunities, and other vehicles for social mobility (Sehlapelo & Terre Blanche, 1996; Taylor & Radford, 1986). It was not until the Employment Equity Act No 55 in 1998 that stipulated the test had to be valid, reliable, fair, and non-biased (Laher & Cockcroft, 2013a: 4). Laher and Cockcroft (2013) explain that the current Employment Equity Act requires “local, emic instruments be developed” and “psychological tests be scientifically reliable, valid and fair” (p. 537). While the Soweto Stress Scale is an emic framework of stressors, it is not a psychiatric assessment that evaluates symptomatology. Nevertheless, the Soweto Stress Scale strongly relates to psychological morbidity (with high internal consistency) and may be a useful primary care screener for those who may need counseling or psychiatric support.

Notably, our “crossover analyses” involved a “data transformation” where we used descriptions from life history narratives to localize a stress checklist, moving from qualitative to quantitative data (Onwuegbuzie et al., 2010, p. 58) and showed relatively high internal validity. We found that the items represented distinct experiences and grouped coherently together in the Soweto Stress Scale. Some items, such as job stress or stress about following doctor’s orders about medical care, and needing to seek medical care, may be excluded from future applications of this scale, but their lack of coherence with the scale as a whole may also reflect the fact that two-thirds of our sample were women, and the majority of our interviews were conducted on weekdays in the middle of the day. With respect to job stress, there was insufficient variation in the responses for the psychometric tests to be conducted.

Limitations

Given the racial, ethnic, and linguistic diversity within South Africa, we do not argue that our “emic” perspective reflects a national scale but rather a very localized one. Since our analysis was from a community-based sample, there may be more medically relevant stressors common to clinical samples. However, despite chronic illness, many people find social stress as stressful as, or even more stressful than, medical problems (Manderson & Warren, 2016). Our findings have specific relevance for patients from Soweto as well as potentially the ethnically diverse and complex peri-urban areas of Johannesburg, such as Alexandra, seeking care at Chris Hani Baragwanath Hospital, and potentially other primary care settings around the city. Given the unique challenges of living in dense urban contexts, the Soweto Stress Scale may be less relevant for people living in rural areas where the local economy is primarily based on subsistence agriculture and animal husbandry. However, we encourage other South African researchers to test this scale and adapt it to local needs in comparable contexts.

Consequently, this scale may not be generalizable and should not be simply applied elsewhere without additional validation efforts. The scale may be used in studies located in other South African townships with residents who have varied ethnic, linguistic, and socioeconomic backgrounds, although there may be other stressors that emerge in contexts without a robust middle-class, which is characteristic of Soweto. We argue emphatically that is not a limitation, but rather the unique benefit of this scale. For instance, the scale may be used by service providers to identify people eligible to access to social and health programs in part because of the compounding effects of life stressors. Moreover, many academic and programmatic researchers conduct research on health, disease, and healing in Soweto; this scale may facilitate a more realistic and localized understanding of how people identify and experience stress, health, and well-being in their everyday lives.

Scale development and adaptation is rarely uniform. People are limited by real constraints, such as time, money, and labor, which can impede locally devised scales. Moreover, ethnographers often do not work with epidemiologists, so using emic perspectives in scales can be difficult without extensive local perspectives. In this case, scale development was built upon life history narratives that provided extensive context about what caused people to experience social distress in Soweto. This point indicates that developing such scales requires a great deal of time as well as collaboration among scholars working between individual-level and population-wide studies. Real constraints can impede the ability to design and implement such scales, too, such as time, money, and labor (Onwuegbuzie et al., 2010; Rasmussen et al., 2015; Weaver & Kaiser, 2015).

Other limitations to this study may relate to the linguistic variation within Soweto and in relation to other areas of South Africa. The 107 interviews utilized to develop the Soweto Stress Scale were conducted in seven different languages, and there often were a mix of two to five languages used between the interviewers, study participants, and, in some cases, translators involved. Moreover, in Soweto, cultural and linguistic localization may influence how people talk about and experience certain stressors; for example, our research team mentioned isiZulu words in Soweto, which sometimes sound very different compared to how they are used or conveyed in KwaZulu-Natal. Thus, the ways in which people spoke about stress may have been conveyed slightly differently across interviews and may have been misunderstood or overlooked without an emphasis on what idioms people used to convey distress. It was for this reason that we focused on specific stressors, except for thinking too much, which was common across interviews. Future studies may identify phrases or expressions of distress, beyond “thinking too much”, and these may add a great deal to our understanding of stress/distress in Soweto.

Conclusions

Moving from ethnography to epidemiology provides a more robust understanding of local drivers of distress. The Soweto Stress Scale shows strong internal consistency and with psychological morbidity, demonstrating that is a relatively useful tool for identifying individuals with more social distress within a community sample. It may also be a useful tool for clinicians to use when screening for social stressors and conditions that may increase one’s risk for psychological morbidity at Chris Hani Baragwanath Hospital, which draws a significant number of patients from Soweto. Similarly, the tool may be very useful for the multitude of clinical care and research conducted at the hospital as well as throughout the community, as it is ethnographically grounded and has broad relevance to the Soweto community in relation to common causes of stress.

Highlights.

• Identified items from ethnography to develop Soweto Stress Scale.

• Locally developed scale of stress can identify those with higher psychological morbidity.

• Used factor analyses to evaluate item relatability and to evaluate the comparative fit index.

• Scale correlated moderately with several dimensions of psychological morbidity.

• Stress scale may not be generalizeable to other contexts without further validation.