1The central research question of this thesis is whether unpaid care and domestic work is a significant constraint to female microenterprise development. This chapter examines the research methodology used to investigate these questions during field research in Luwero District, Uganda. It begins by outlining the specific research tools that were used, providing justification for their appropriateness and inclusion. Next, it details the sampling procedure employed to collect representative sample data from the target population. After establishing how the data was collected, the third section describes how the data was analysed. The final section considers the limitations of the chosen research methodology.

Research Tools

2The choice of research tools for this study were determined by their ability to respond to the specific sub-questions of the research problem: how much unpaid work is being carried out by women micro-entrepreneurs; what are their self-reported constraints to microenterprise development; and how do social norms on gender influence the allocation of unpaid work?

3The first sub-question required an accurate understanding of each respondent’s time allocation, while the second and third sub-questions required eliciting the opinions of the women. A time-use survey was therefore chosen as the appropriate research tool for the former, while a questionnaire was used for the latter. Throughout the research process, field notes were also taken. This section outlines each of these three tools, largely focusing on the time-use survey, given its relative obscurity as a research method.

Time-Use Survey

4A time-use survey is a research tool used to analyse how people allocate their time. Specifically, it measures the amount of time dedicated to different activities within a given period to provide detailed and comprehensive information on how individuals spend their time (Hirway 2017). Given the difficulty in accurately measuring unpaid work through simple self-reported measures, as outlined in the Literature Review, the time-use survey appears to be a more reliable method to calculate this time.

5The methodological approaches to using time-use surveys vary considerably. The key differences are summarised here, along with a justification for the decisions made in this regard for the field research. Overall, time-use surveys can take one of two forms: a diary approach, or a stylised approach. The diary approach asks respondents what they did during each period within a 24-hour day. These results are then aggregated and classified into different activities to provide a quantitative summary of how individuals allocate their time. The stylised approach, on the other hand, consists of a set of questions asking respondents how long they spent in total on specific activities over a period of 24 hours (Budlender 2010). Therefore, the diary approach might ask respondents: “What did you do from 09:00 to 09:30 yesterday morning?” while the stylised approach might ask: “How much time did you spend yesterday on the unpaid care of your children?”.

6The benefit of the diary approach is that it has the potential to yield extremely rich data. By walking respondents through their day, it is likely to provide a more accurate measurement of the time spent on different activities, compared to simple self-reported measures. Beyond establishing how much time is spent on activities, it also tells the researcher when in the day these activities take place. On the other hand, the data collection process for the diary approach is labour-intensive and requires the ex-post classification of activities. The benefit of the stylised approach is that it is much simpler, reducing the burden on both the respondent and the researcher. The questions can directly elicit the amount of time spent on certain activities without further coding required. However, self-reported measures from the stylised approach are likely to produce less accurate results compared to those of the diary approach (Budlender 2007). Given the importance of collecting detailed and accurate time-use data for this study, the decision was therefore taken to proceed with the diary approach.

7Another benefit of the diary approach is that it allows for the inclusion of simultaneous activities, which is particularly relevant for the study of unpaid work – as alluded to in the Literature Review chapter. Simultaneous activities can be captured through a second column in the diary, which also encourages respondents to think in more detail about how their time is spent. Where two activities occur simultaneously, they can either be ranked according to the primary and secondary activity or given equal weight (Budlender 2010). For the field research, a second column was added to the time-diary to include simultaneous activities. Respondents were actively encouraged to consider if they were doing anything else at the same time as their initial response for each period. To reduce the total time burden for the respondent and the researcher, simultaneous activities were given an equal weight – with each time period divided equally between activities.

8Context variables can provide complementary information to time-use data in order to provide a richer understanding of the allocation of people’s time. Examples include where the activity is taking place; who is present alongside the respondent; for whom the activity is being carried out; and whether the activity is paid or unpaid (Hirway 2017). While most of this information would be interesting to collect, and could perhaps contribute additional insight to the findings, the significant time burden seems to outweigh the usefulness of this data – with the exception of information on whether the activity is paid or unpaid, given the focus of this study. Therefore, while other context variables were not considered in this time-use survey, respondents were asked to explicitly state when they were carrying out activities for payment.

9Conducting time-use surveys over multiple days and with multiple household members would have yielded more detailed data – but would have also substantially increased the burden on the researcher and the households under investigation. Similarly, if using very small (e.g. 10-minute) periods for the diary would have produced more granular data on time-use, it would have also been extremely labour-intensive and potentially could also have induced satisficing in respondents. Overall, decisions on the methodological approach for this study were made in order to respond to the research question as fully as possible given time and budget constraints. Further considerations are discussed in the Limitations section later in the chapter.

10The time-use survey used a diary approach to obtain information on the time allocation of the women micro-entrepreneurs. They were asked by the researcher to recall their previous 24 hours, after confirming that it had been a regular working day for them. The researcher used a time-use survey with 30-minute periods as a template, filling it in by hand according to the responses. Follow-up questions were asked, to establish whether activities were done for payment and whether other activities were being done at the same time. The finalised time-use survey drawn up for the field research is shown in Appendix I.

Questionnaire

11The second element of the research consisted of a questionnaire. This tool was used to establish women micro-entrepreneurs’ self-reported constraints to developing their business, and how social norms on gender influence the allocation of unpaid work within their community. The questionnaire also elicited additional information at the individual, business and household level to provide a better understanding of the background characteristics of the sample and observe how these factors interact with the rest of the data.

12The questions on constraints to enterprise development focused initially on all possible financial and non-financial barriers, before explicitly asking about time constraints and unpaid work. Social norms were estimated through a series of attitudinal questions asking the women about their own opinions and about their beliefs on the opinions of others. Finally, the background information questions asked basic questions about the women, their business and their household. Other questions were added after consultation with the wider research team in order to provide relevant information specific to their project. The questions most relevant in addressing the research problem of this study are explored in greater detail in the Results and Analysis chapter.

13The format of the questionnaire was mostly multiple-choice questions. This decision was taken to minimise the burden on the respondents and the researcher, given the volume of questions involved, simplifying the data collection process and allowing a greater number of respondents to be sampled. The finalised choice-set of responses was obtained through an iterative process with the wider research team and local BRAC staff, each of whom brought their unique experience and expertise to improve the options available. For instance, one question asked the women about the mode of transportation commonly used to get to their business but failed to capture the local knowledge that many people travel by boda-boda, or motorcycle taxi. A few questions had follow-up, open responses. This allowed qualitative comments to be considered for the questions deemed most important to understanding the constraints faced by these women to growing their business, and what could be done to alleviate them.

14The questionnaire was tested on members of the BRAC research team, with questions rephrased and restructured based on their feedback. It was then coded into programming language for SurveyCTO, a mobile data-collection tool, and loaded onto mobile devices to be used in the field for data-entry. The interviews with the women micro-entrepreneurs were conducted in Luganda, the local language, by the research assistant from BRAC. Responses were then translated into English and noted down accordingly. The interviews generally took between 40 and 60 minutes per respondent, inclusive of the time-use survey. The finalised questionnaire used for the field research, together with the time-use survey, can be found in Appendix I.

Field Notes

15Field notes were taken over the course of the research visit. These consisted of observations, records of informal conversations and personal reflections. The rationale for incorporating these field notes into the research is to provide a richer description and understanding of the observed phenomena.

Sampling Procedure

16This section outlines the sampling procedure chosen for the field research. It begins by restating the target population and the research setting, before detailing the sampling strategy employed. Difficulties during field research led to this strategy being somewhat adapted, which is openly discussed. An overview of the final sample concludes the section.

17The target population of this study was female BRAC microcredit clients in Luwero District, Uganda, with children aged three to five. The rough census carried out as part of this research returned an estimated total target population of 1,315 women. Using this census data, the research sample was selected from the target population using a multi-stage stratified random sampling technique, which can be understood in three steps. First, the total target population was divided into three, according to the BRAC microcredit branch they belonged to: Bombo, Luwero or Wobulenzi. Next, microcredit groups – consisting of roughly 10-20 members – were randomly selected from each of these branches. Finally, within these randomly selected groups, respondents were randomly chosen from among those members who fit the selection criteria. Overall, this sampling strategy was chosen for its ability to provide data on a wide range of respondents from different microcredit groups across all three branches within the district. This was therefore likely to yield a representative sample, accurately reflecting the wider population, meaning that findings from the study could be generalised to the larger, target population.

18While this sampling strategy appeared straightforward in theory, the reality was more complex. The census data did not provide the precise location of the women’s businesses, so it was not possible to visit the women at their workplace unannounced and conduct the interviews. While most women provided a contact telephone number as part of the census, they were often unreachable. Where contact was established, the time constraints of these women meant that many of them said they were unavailable – not surprising given the scope of this research. This produced two key restrictions to the collection of representative sample data. Firstly, only those women who agreed to make time for the interview would be sampled, leading to a strong bias towards those who are less time-constrained. Secondly, this method was time-consuming and expensive for the researcher as it entailed constantly travelling to different locations to conduct interviews with available women. The greater time and expense required per interview using this technique meant that the total sample size would be considerably lower.

19To ensure a sufficiently representative sample size and a more efficient method of data collection, an element of convenience sampling was introduced. The three BRAC microcredit branches were each supported by a team of credit officers, each of whom was responsible for overseeing a number of microcredit groups: dispensing loans and collecting repayments. Groups tended to meet on a weekly basis at a suitable location close to the members, with credit officers attending multiple group meetings per day. A decision was therefore made early on in the data-gathering process to visit the local microcredit branch each morning and follow one of the present credit officers throughout their day of group meetings. At each group meeting, eligible respondents would then be identified and randomly selected to be interviewed. Thus, respondents were still chosen from all three branches, and eligible members from the visited groups were still selected at random. The only non-random element introduced was the group selection, as this was determined by the credit officer. Overall, the addition of a convenience element to the sampling procedure dramatically reduced the time and expense of the field research without sacrificing the representativeness of the total sample – and from a practical perspective probably improved it.

20The total sample of women micro-entrepreneurs interviewed was 64, close to five per cent of the total target population. It transpired over the course of the interviews that part of the selection criteria had been translated as having caring responsibilities for a three to five-year-old. This led to the inclusion of grandmothers as eligible respondents where these women were responsible for looking after their young grandchildren. Given that these women were still micro-entrepreneurs with significant caring responsibilities, they were included in the final sample. Out of the 64 women, six reported that their previous day had not been a regular working day. This meant that their combined questionnaire and time-use data could not be used, and these women were dropped from the final sample. It also emerged during one interview that the respondent did not actually have a child between three and five years old, so this woman was also dropped. This resulted in a final sample size of 57, equating to 4.3% of the total estimated target population.

Analysis

21The time-use survey, questionnaire and field notes yielded a mixture of quantitative and qualitative data. This section explains how this data was used: starting with the coding of data for quantitative assessment before turning to the data-analysing process undertaken for both sets of data. It concludes by demonstrating how the mixed methods approach, combining quantitative and qualitative data, likely provided a more detailed understanding of the relationship between unpaid work and female microenterprise development.

22Coding of the time-use survey and questionnaire was undertaken in order to facilitate the data analysis process. For the questionnaire, coding was relatively straightforward and consisted of assigning logical numerical values to the various sets of possible responses to questions. SurveyCTO, the mobile data-collection tool used for the questionnaire, digitally documented responses in a spreadsheet format – which allowed for simple ex-post coding using Microsoft Excel. Coding the time-use survey was more complex. Each of these was filled out by hand and therefore required digitising. Moreover, each activity – described in words in the time-use surveys – required numerical classification.

23The United Nations Statistical Division publishes an International Classification of Activities for Time Use Statistics (ICATUS), which provides a framework of standardised measures of time allocation. The most recent iteration, ICATUS 2016, is a three-level hierarchical classification of activities, with well-described, mutually-exclusive and exhaustive categories (UNSD 2017). The first level of this classification consists of nine major divisions, shown in Table 1. A more detailed classification of the ICATUS 2016 can be found in Appendix II.

24A simple typology of work was constructed based on these major divisions. Paid work corresponded to employment and related activities. Unpaid work incorporated production of goods for own final use; unpaid domestic services for household and family members; unpaid caregiving services for household and family members; and unpaid volunteer, trainee and other unpaid work. These activities coincide with the definition of unpaid work outlined in the Literature Review chapter. The remaining activities were classified as non-work.

25The major divisions provided a sufficient level of detail for this research and were therefore adopted as the framework for classifying and analysing the time-use data. The only exception was the addition of sleep, a sub-division of self-care and maintenance, as a category of its own in order to specifically analyse the time-allocation of waking hours. Each physical time-use survey was analysed by the researcher, with activities aggregated according to these ten categories and coded accordingly. This data was then converted to a digital spreadsheet using Microsoft Excel.

26The quantitative data generated from the questionnaire and time-use survey were analysed using a combination of Microsoft Excel and Stata, a statistical analysis programme. Manipulation of the data included measures of central tendency (mean, median, mode), disaggregating the data according to certain characteristics, conducting t-tests to establish if two subgroups within the dataset have the same mean value for a given variable, and running basic regression analysis. The quantitative data from the time-use survey generated an overview of how much time was allocated to different activities within a 24-hour period. A qualitative examination of the time-use surveys, on the other hand, indicated when specific activities occurred as well as their simultaneity. The qualitative data from the time-use surveys, along with the open responses from the questionnaires and the field notes were collated by the researcher and analysed for common themes across respondents.

27This study used a mixed methods approach, triangulating the quantitative and qualitative data to provide different perspectives on the same research question (Denzin 1970). The decision to use this approach was based on the complementary nature of the two sets of data, which would hopefully lead to greater validity of any inferences made from the sample data and provide a richer understanding of the complex relationship between unpaid work and female microenterprise development.

Limitations

28The limitations of this research study were mostly a function of time and budget constraints. This section examines the key limitations of the time-use survey, the questionnaire and the broader fieldwork to demonstrate awareness of the limits of this research and explain why the specific methodology outlined above was pursued.

29The time-use survey was conducted with women micro-entrepreneurs from a specific location (Luwero District) at a specific point in time (Spring 2018). The time-use statistics therefore do not account for geographical or seasonal variation, which could significantly influence time allocation – particularly in a predominantly agricultural area such as Luwero. However, given the time and budget constraints of the researcher, it was neither feasible to return to this research site on multiple occasions throughout the year, nor conduct additional fieldwork of similar quality in other districts. The chosen research method provided time-use data over the period of a single day for each woman micro-entrepreneur, giving a snapshot of their time allocation on a regular working day. A more accurate way to estimate the lived realities of these women would be to conduct a time-use survey with each of them over the course of multiple days. With the limited resources available, and acknowledging the trade-offs involved in these alternative approaches, it was deemed better to concentrate on collecting as many data points as possible from different women within a single district – thereby providing as representative a sample as possible from this smaller target population.

30A second limitation of the time-use survey is that it relied upon respondents accurately recalling how they allocated their time. While the diary approach of walking the respondent through their previous day appears better than the stylised approach in this regard, it is nonetheless prone to reporting errors. A more accurate measurement of time-use could have been achieved through participant observation. Three reasons justified the original, face-to-face recall approach. Firstly, it would have been prohibitively time-consuming to spend a full day with each respondent. Second, issues around gender dynamics would have had to be addressed, including the cultural sensitivity towards an Asian man spending time in the households of single and cohabiting African women with young children. Finally, respondents might have altered their behaviour if they knew they were being observed, commonly known as the Hawthorne effect. Given these restrictions, the face-to-face recall method was retained as the optimal approach with the resources available.

31A final limitation of the time-use survey is that it failed to fully capture the temporal demands of unpaid work. This is because unpaid work is not merely an activity, but also a responsibility. It constrains the allocation of time even when no unpaid work is being conducted (Folbre 2014). Therefore, by only calculating time spent on specific activities, it is likely that the time-use survey underestimated the true demands of unpaid work on these women. However, combining the analysis of the time-use data with complementary, qualitative comments from the women could overcome this problem by providing a fuller understanding of the demands of unpaid work on the time of these women.

32One key limitation of the questionnaire is that it tended to use a five-point scale for attitudinal questions, including a neutral, middle option. Psychological insights into the behaviour of survey respondents suggest that the inclusion of this neutral option encourages satisficing (Pasek and Krosnick 2010). However, the trade-off faced when removing this option is that it limits the full choice-set of responses. In the interests of not obliging the women to choose either way when they may have had neutral feelings towards the question being asked, the middle option was retained.

33Overall, the field research may have been limited by two factors. Firstly, the sample size was small relative to the target population, which meant that the findings may not have been representative. The sampling strategy somewhat addressed this through the stratification and randomisation procedures. Moreover, while the low statistical power from the small sample size meant that relationships between variables were harder to detect, those associations found to be statistically significant would carry substantial weight. Secondly, the research was conducted only with women micro-entrepreneurs. It was therefore not possible to compare the time-use of similar males and females to show that women carried out more unpaid work than men. This therefore remains a large assumption of this study, though the phenomenon is well-documented in the literature and is corroborated by the questionnaire data. Similarly, the attitudes of men towards social norms on gender are estimated through the responses given by the women. Ideally, the time-use and attitudes of men would have been directly elicited; however, the limited available time and budget meant that this was not feasible.

34Finally, it is worth noting two key points regarding the scope of this research. Firstly, it is not possible to establish causality from the data collected. Any relationships that appear to exist between variables in the data can only be evaluated as simple associations. Suggestive evidence from the qualitative data could allow for an educated guess as to the mechanism behind this relationship and the direction of causality – but this remains a hypothesis to be tested and cannot be proven with this data. Secondly, the data collected as part of this study is highly context-specific. The sample consisted of women micro-entrepreneurs living in Luwero District, Uganda, who were BRAC microcredit clients and cared for children between three and five years old. The findings documented in the next chapter should therefore be viewed within this context.