The Welfare, Children, and Families: A Three-City Study (hereafter, the Three-City Study) is a longitudinal examination of low-income families from Chicago, Boston, and San Antonio. With its large sample (n = 256) and two years of ethnographic data collection for each family, the study promised to provide the first two authors (Speirs and Vesely; hereafter referred to as we) a unique opportunity to understand childcare instability within the context of families’ lives and larger policy structures. Our excitement about using this dataset was tempered, however, when we were confronted with thousands of pages of interview and fieldnote data, summary documents, and confusing participant identification numbers and file names. We very quickly found ourselves lost in the data. Here we tell the story of how we found our way out.
This chapter begins with a brief description of the Three-City Study dataset and our findings concerning childcare instability and transitions. We then detail the strategies that we used to become familiar with a large-scale qualitative dataset, select an analytic sample, and code the data. We end the chapter with reflections on analyzing a large-scale qualitative dataset.
The Three-City Study was designed to explore the impact of welfare reform on children and families; a detailed description can be found at http://web.jhu.edu/threecitystudy. For the project described in this chapter (Speirs, Vesely, & Roy, 2015), we used the study’s ethnographic component, which involved 256 families. For 12 to 18 months ethnographers visited each family at least once a month to conduct semi-structured interviews and participant observations during everyday events at the family’s home, doctor visits, and appointments with social workers. Semi-structured interviews were conducted using 10 required topical interview guides (e.g., daily routine, welfare and employment experiences) and 10 optional protocols (e.g., childcare and father involvement). Interviews were transcribed verbatim. Observations were recorded as detailed fieldnotes that also included the ethnographer’s insights. Data collection is described in detail in our original report (Speirs et al., 2015). Given the prolonged engagement and frequent contact, many families had at least 20 data collection points and hundreds of pages of data. When first opening these files we were overwhelmed by the amount of data for each family and the prospect of attempting to read through them multiple times, but also excited about the level of detail this much data could provide.
Speirs and Vesely, who led the secondary data analysis described in this chapter, were not involved in data collection and had not used the dataset before this project. The third author, Roy, was a lead ethnographer for the Three-City Study, coordinated data collection for the Chicago site, and had published from the dataset (e.g., Roy & Burton, 2007; Roy, Tubbs, & Burton, 2004). Roy worked with Speirs and Vesely (as doctoral students) to navigate the complicated set of study documents and brainstorm preliminary analyses, and then handed over primary data analysis to them so that they could carve out a topic of shared interest.
We identified four types of stability and instability (Speirs et al., 2015). Mothers who organized planned transitions made changes to their care arrangements that were planned in advance and supportive of their families’ well-being. Averted transitions were the work mothers did to maintain functional care arrangements that were supportive of their child’s healthy development and/or their own ability to maintain employment. Mothers who experienced forced transitions had to leave a preferred care arrangement. Failed transitions occurred when mothers were not satisfied with their care arrangement but were unable to secure a new one. We also detailed the factors that led to each of these types of transitions. For example, we found that mothers were able to make planned transitions when they had time (either by being able to anticipate transitions or having reliable and flexible secondary care arrangements); organizational and planning skills; and support from family, friends, and caseworkers.
Most qualitative researchers become familiar with their dataset and begin analysis by writing memos during data collection. This was not possible for our project as we did not collect the data. Instead, we reviewed summary documents created by the original study team to familiarize ourselves with the dataset and begin analysis.
Timelines for each family that graphically illustrated changes in several areas (e.g., employment, income, childcare arrangements, health conditions, and housing) were the first summary documents that we used (see Figure 9.1). We received the timelines as hard copies printed in color on legal-sized paper and as electronic files (one per family) in Microsoft Excel. The timelines allowed us to quickly orient ourselves to each family by providing an overview of every time a family transitioned from one care arrangement to another as well as how changes in one life domain (e.g., employment, childcare, housing) preceded or followed changes in another. We could immediately appreciate the complexity of families’ lives with this concise visual tool.
The second summary documents, multi-page (range: 20–80 pages) family profiles, were developed by ethnographers at the main Penn State office during data collection to organize chunks of relevant text from fieldnotes and interview transcripts into one Microsoft Word document. These family profiles were a critical “way-in” to the data and served as first-pass summaries of the entire set of data for each family.
Family profiles included four sections: (a) enrollment and study information (e.g., participant ID and study entry date); (b) participant background information (e.g., names and descriptions of family members); (c) data collection details (e.g., date data were collected, file names, and interview or fieldnote content); and (d) detailed notes from each data collection time point. The fourth section was the essence of the profile. It began with a description of how the family was recruited to the study, their home, neighborhood, and their social network. Most importantly, there were summaries of the relevant information from each interview or fieldnote organized by topic. See Table 9.1 for an excerpt from the childcare section of a profile.
The length and complexity of the family profiles varied depending on how long the family was enrolled in the study and how many interviews and observations they had completed. For the full sample we had thousands of pages of family profiles.
We began familiarizing ourselves with the dataset by noting major events on each family’s timeline, and then reading their family profile to flesh out the details. We started with the childcare section of the timelines and family profiles and then read other related sections (e.g., employment, daily routines). We had anticipated strong links between employment and childcare, but were also able to see how health issues and housing instability related to care arrangements. While reviewing the summary documents, we each took individual notes concerning potential study aims and how different aspects of families’ lives were associated with childcare. We made these notes in Word documents that we each kept to record questions, ideas, and memos. At this point our notes included what was not available in the timelines and family profiles (e.g., detailed information about mothers’ motivations for childcare transitions) and that data were uneven across families (e.g., some families did not complete a childcare interview).
|
[Fieldnote Date and File Name] When Lorene was working at McDonald’s, Jordan’s father (Adam) watched Jordan. He also had a job and scheduling was difficult. Sometimes Jordan was watched by his grandmother but it did not work well. Lorene could not find childcare and had to quit her job. [Interview Date and File Name] Lorene is upset with her daycare center. It is inadequately staffed. Jordan developed a rash because the staff did not change his diapers. Lorene did not say anything. Recently, it looked like Jordan was bitten and the staff tried to hide it. Lorene is going to take Jordan to the doctor and if it is a bite, she is going to say something to the daycare. Lorene says, “They can sanction me, take what they want, but they have to give me a chance to get another daycare and do an investigation.” |
|
[Fieldnote Date and File Name] Adam and Lorene broke up and Lorene was left without his help which impacted her childcare situation. She had to miss work. [Fieldnote Date and File Name] Lorene had her children in a state-sponsored daycare, but was not able to meet the participation requirements of the Texas Workforce Commission and lost this daycare. |
Note. All names are pseudonyms.
After each reading 10 families’ timelines and profiles, we began meeting with Roy two to four times a month to discuss our notes, overall impressions of the data, and potential study aims. One note that Speirs wrote early on in this process concerned differences across types of care: “Should we look at the kind of care (center, kin care, home-based) mothers switch to and from? Is it easier to switch to or from one kind? What are the complications with switching to and from different kinds of care?” (Speirs, memo, 2014). The three of us discussed these kinds of questions at our meetings, came to consensus (or agreed to put off making a decision), and used that decision to guide our subsequent reading of the data. We continued to move through the dataset in this manner: reading timelines and family profiles for a few families, then meeting to discuss our initial ideas about the data. We also began to review the literature on childcare stability (e.g., Adams & Rohacek, 2010; Chaudry, 2004) and decision making (e.g., Van Horn, Ramey, Mulvihill, & Newell, 2001) to have a sense of the questions other researchers were addressing and how our data could add to this conversation.
We decided to focus on mothers’ transitions between care arrangements and how and why they made these transitions. Our literature review had suggested that there was limited research on childcare transitions and instability and the extant research was largely quantitative. We thought we could use the Three-City Study ethnographic data to bring to light mothers’ perspectives and lived experiences as well as generate findings that might inform efforts to develop policies and programs to help mothers find supportive childcare.
We decided an important first step would be to count the number of childcare transitions for each family and document the reasons for them. We wanted to be sure we were capturing all of the childcare transitions that we had data for, not just those that were the most interesting or frequently discussed. We also attempted to document the number of school and work transitions in order to explore relationships between childcare, work, and school. We created a spreadsheet to organize this work (see Figure 9.2) and stored it in a Dropbox folder so that all three of us could view and edit the same spreadsheet.
Although some of this information was already summarized on the timelines and in the family profiles, creating our own summary spreadsheet proved a critical step in developing our familiarity with the dataset and “owning” the analyses. In putting together the summary spreadsheet we were forced to identify which aspects of the family’s lives were important to our analyses (childcare, school, and work) and dig through the available documents to find as much information about these transitions as we could or acknowledge where data were missing. This process helped us move from reading through the original study team’s summaries and ideas about the data, which were often included in the family profiles and interview transcripts, to developing our own insights into families’ childcare stories and drawing conclusions about childcare stability. Engaging with the dataset in this way helped us begin to take ownership of the data and our analyses.
As we put together this spreadsheet, we started to define different kinds of transitions. In looking at the motivation for each care transition, we realized that some were out of the mothers’ control (e.g., a center unexpectedly closing). We recorded our ideas about the different kinds of transitions as notes and memos in Word documents that we kept separately on our own computers. Speirs, Vesely, and Roy also discussed the different transition types during hour-long meetings, held once or twice a month. At least one of us would take notes by hand. After the meeting one or both of us would update our electronic notes to reflect any new ideas generated at the meeting. We were probably not as systematic in this note keeping as we should have been. There were times we forgot to update our electronic notes and instead relied on our memories and handwritten notes. We likely lost details from some of our discussions. When we did write notes, they included a description of the transition (e.g., some mothers make plans to move their child from home-based care to a preschool to prepare them for school) and examples from the interviews or family profiles illustrating the transition.
While we were creating our summary spreadsheet, we continued to read the literature to be able to situate our analyses within what was already known about childcare transitions and instability. As we read the literature, we deposited relevant articles as pdf files into our Dropbox folder. We compiled article summaries in a shared spreadsheet that included one row for each article and columns for the research question and hypotheses, dataset or sample, definitions of stability or transitions, type of analysis, results, and notes about (among other things) how the article was relevant to our study. If we found something particularly interesting, we shared it over email or at our meetings.
In reading the literature we were influenced by two discoveries: (a) the distinction between changes (predictable planned childcare transitions) and (b) instability (unanticipated changes; Lowe, Weisner, & Geis, 2003). We had seen this distinction in our data and noted that although some mothers were able to plan transitions, others (or sometimes the same mothers) were left to respond to unanticipated or unwelcomed changes. We came to think of this second transition type as being forced on the mothers.
Our second discovery was that many quantitative studies (e.g., Pilarz & Hill, 2014; Tran & Winsler, 2011) assumed that childcare stability would be associated with positive child outcomes and instability with negative outcomes. However, we noted that there were several mothers in our sample who desperately wanted to change their childcare arrangement but could not find a suitable alternative. These “failed transitions” were impossible to ignore because of the psychological toll that some mothers described. These cases seemed to challenge the assumption in the quantitative literature that childcare stability should be associated with positive outcomes.
Having identified three kinds of transitions (planned, forced, and failed), we explored how they were related. We realized that they could be defined by two dimensions, whether they represented stability (failed transitions) or instability (planned and forced transitions) and whether they should be supportive of child and parent well-being (planned transitions) or not (forced and failed transitions). Having defined three cells in a two-by-two matrix (Table 9.2), we turned our attention to the fourth cell—stability that was supportive of well-being. We began looking for examples and realized that many of the mothers in our sample worked hard to maintain childcare arrangements that provided safe and enriching environments for their children and allowed the mothers to work or attend school.
|
Childcare instability |
Childcare stability |
|
|---|---|---|
|
Supportive of general well-being |
Planned transition |
Averted transition |
|
Unsupportive of general well-being |
Forced transition |
Failed transition |
Note. From Speirs et al. (2015). Reprinted with permission of Elsevier.
Having defined four types of transitions using the timelines and family profiles, we wanted to use the full interview transcripts and fieldnotes to confirm that we had an exhaustive list of transition types and that they were relevant for large proportions of the sample. However, we were overwhelmed by the thought of coding multiple transcripts and fieldnotes for 256 families. Luckily, around this time we traveled from Maryland to North Carolina to meet in person with Linda Burton, the principal investigator of the Three-City Study ethnography, and she suggested selecting a smaller analytic sample so as not to lose depth and detail by attempting to analyze the full sample (L. Burton, personal communication, June 26, 2012).
Guided by our research question, we drew an analytic sample of 36 families. In order to preserve the purposive sampling built into the original dataset, we selected 12 families from each of the three cities and race/ethnic groups (African American, Latino, and non-Latino White) sampled from for the original study. Our original report includes a detailed description of sample selection (Speirs et al., 2015).
We also limited the number of interviews that we coded. For the families in our analytic sample we had access to 10–20 interviews conducted by Three-City Study ethnographers, as described in the second section of this chapter. Transcripts for each interview ranged from five to 35 single-spaced pages. We quickly realized that not all of these interviews would be useful as not all of them addressed childcare. We found it most useful to focus on the childcare, employment, and family routines interviews. Skimming by hand and text searching (using the find function in Microsoft Word) for keywords such as “childcare” or “work” in the remaining interviews indicated that we should also read the support network interviews as they often included details about kin care and secondary care arrangements. We also continued to use the family profiles to gain an overall portrait of participants’ lives. The contextual details found in the family profiles (e.g., who was in the household and how money was allocated at different points in the month) helped us to interpret how each transition unfolded.
After determining that we would focus on the childcare, employment, family routines, and support network interviews, we divided the 36 families in our sub-sample between the two of us. Speirs took the Chicago families, Vesely the Boston families, and the families from San Antonio were divided between us. We each uploaded the interview transcripts, fieldnotes, and family profiles (all of which were Word documents) for 18 families into ATLAS.ti software. We had one file per interview, fieldnote, or profile. We then began coding in order to confirm the four transition types we had identified. As we coded, we moved between summary documents and full interview transcripts. Before coding an interview, we reviewed the timeline and family profile to orient ourselves to each mother’s major childcare and employment transitions. We would then code the transcripts and once again consult the summary documents to ensure we had not missed any important transitions. We used ATLAS.ti to code any section of interview text that described a childcare transition as a “planned,” “forced,” “failed,” or “averted” transition, or as “unsure how to code” (see Figure 9.3).
We worked through the sections of text coded “unsure how to code” one at a time. Some were easily resolved after reviewing the text a second time with fresh eyes. Others required discussions at our in-person meetings. For example, we were initially unsure if the following text described a planned or forced transition.
When I moved down here with (my husband), I supported him. [laughs] While I was working, he was with (target child). I didn’t have no problems with it. But then, I got kind of pissed off about it, cause I got kind of tired of going to school and working at the same time while he was sitting at home, watching (target child). I didn’t consider that 50/50. Until after, I kind of thought about it and was like, well, wait a minute. I do need a sitter. Who’s going to watch the child for free, besides his own daddy? But then, at that point, he was sick and in the hospital and I kind of regretted that. Getting into that little argument with him about that. But—I let him know I loved him, and after he got well then he could come back.
We debated if this was (a) a planned transition because this mother wanted her husband to work rather than care for their child and seemed to take action to make this happen, or (b) a forced transition because she eventually saw having her husband care for her child as a beneficial arrangement, but was forced to find another caregiver after her husband became sick. After searching the rest of the interview transcripts and the family profile, we found no evidence that the father had actually stopped caring for the child after the mother mentioned that she was not happy with this arrangement but before he became ill. We therefore decided that this was a forced transition. Additionally, the mother’s statement that “after he got well then he could come back” suggested that she was ultimately happy with the father caring for their child and was forced to find another caregiver after he was hospitalized.
After coding all transitions as planned, failed, forced, or averted, we used ATLAS.ti to produce four reports. Each report listed all of the quotes associated with one transition type (i.e., all sections of interview or fieldnote text that had been coded as a planned transition). We each took responsibility for two transition types: Speirs coded failed and forced transitions and Vesely planned and averted. We shared the reports so that we each had all of the quotes associated with our two transition types. We independently uploaded these reports into ATLAS.ti and began reading through them to understand how and why each transition happened. We focused on factors that promoted transitions because one of our goals with this paper was to provide information that could inform policy and program development.
We moved through three stages of coding: open, axial, and selective (LaRossa, 2005) as described in our original report (Speirs et al., 2015 and Figure 9.4). We began by identifying questions that would shed light on how and why transitions happened, then created open codes that answered these questions. For example, for the code “failed transitions,” one question we were interested in was why mothers wanted to transition. We applied open codes that described their motivations for attempting to change care providers and then grouped these codes into three categories (or axial codes). “Concern about quality or safety of care arrangement” was for codes related to mothers’ concerns about health and safety or the caregiver’s ability to provide appropriate care. “Logistics” included codes for inconvenient operating hours for centers, provider not available at agreed on times or dates for kin care; and inconvenient location. “Child missing an important experience” was for mothers’ concern about their child missing opportunities to promote social-emotional or cognitive development.
In reading through the quotes associated with the “Child missing an important experience” code, Speirs noted that one mother responded to not being able to enroll her daughter in a preschool program by teaching her some of the skills (e.g., writing her name, counting) that she would have learned in preschool. She wrote the following memo: “This mother has found another way to provide these experiences for her child. Suggests resilience and planfulness even in the context of a failed transition. Are other mothers doing this?” (Speirs, memo, 2014). During the next meeting, Vesely confirmed that she had also seen examples of this. After the meeting, Speirs went back to the data and coded additional examples. Looking for these examples and thinking more about how mothers responded to failed transitions lead to the realization that an important dimension of failed transitions was that some seemed to have a greater impact than others (e.g., using an unsafe care arrangement was more distressful than using an inconveniently located one).
We coded independently and then met to talk through our findings. At this point in the project, we were living in different parts of the country so we held these meetings over the phone. One person would present her conclusions and the other would support or challenge them. At one point, Vesely suggested that having at least one secondary care provider could allow mothers to make planned transitions by providing emergency care and time to carefully search for a new provider when a primary provider was suddenly and permanently unavailable. Speirs expanded on this idea by offering examples where short-term emergency care from a secondary care provider meant a mother did not have to find a new primary care provider if her current one was unavailable for a short period of time (e.g., during a vacation). We fed off each other’s excitement. If one person suggested a theory or conclusion about the data, the other would often jump in with examples supporting it. However, we were not shy about challenging each other’s conclusions. At this point, we had worked together for several years and had a great deal of mutual respect. We were quick to introduce examples that contradicted each other’s conclusions and speak up when we thought the other person was incorrectly interpreting the data. At this point in the coding process, Roy was not joining our phone calls, but was available when we needed to run something by him or for general guidance.
We took different approaches to organizing and recording our open and axial coding. Vesely loaded the interview transcripts into ATLAS.ti and applied an initial set of open codes. She then switched to a Word document for axial coding by creating documents that listed all of the quotes associated with each code. As her standard analysis process, Vesely stores data and conducts initial analyses in ATLAS.ti so that she has an electronic record that includes all of the codes and their definitions. This can be easily updated and used to aggregate data by code. Vesely has gotten into the habit of using Word documents during axial coding, as she typically works on large research teams that include community members who may not have access to ATLAS.ti, and she has experienced difficulty sharing ATLAS.ti projects with other researchers especially across PC and Mac operating systems.
In contrast, Speirs prefers coding entirely in ATLAS.ti because it offers two features that she has come to rely on. First, ATLAS.ti provides a window where code memos can be written and easily edited (see Figure 9.3). She uses this window to write definitions for each code and then refers back to that definition to ensure that her understanding of the code has not changed. If it has changed, she either widens the definition or creates new codes. Second, she likes the ability to quickly see all of the quotes associated with a code. During and after an initial coding pass she will look at all of a quote’s associated codes to ensure that all quotes describe the same phenomenon and that additional codes do not need to be added.
Selective coding came after we had completed open and axial coding and had begun writing the findings section of our manuscript. We decided that the central theme organizing our findings was the idea that both stability and instability could be positive or negative. This was the central theme that all of our open and axial codes spoke to and the largest take-away message from the paper.
When we began a secondary data analysis of the Three-City Study ethnography, we did not realize how different it would be to work with a dataset of this size than it had been to work with smaller datasets. We quickly realized that this project would require creative analytic strategies. In the following, we reflect on lessons learned and effective approaches, and provide cautions for researchers embarking on similar analyses.
Throughout this project we navigated a tension between using the full potential of the dataset and being realistic about the amount of data we could analyze. At two points we made decisions to limit the data that we would use: we selected an analytic sample of 36 families from the full sample of 256 and focused on four of the 10 to 20 interviews available for each family. It would not have been realistic for two people to read thousands of pages of interview transcripts and fieldnotes multiple times to analyze the full dataset. Using the full dataset would have meant relying more heavily on summary documents, skimming interview transcripts, and using text searches to find relevant passages, ultimately losing the depth of detail that two years of data collection allowed. In limiting the dataset to 36 families we were able to follow a childcare transition across multiple interviews and observe how it started as an idea or a mother’s desire to make a change, became a plan, met roadblocks, and either resulted in a transition or the mother starting over. We were able to track down passing references to conflicts with childcare providers or trace out how a family feud meant a mother lost her most reliable source of childcare when her mother stopped talking to her.
These decisions to limit the sample size and number of interviews were made with careful consideration. We fought the idea that we could not use the full sample for some time and were only convinced to do so because Linda Burton (a researcher we greatly respected) suggested it. We had internalized the idea that a larger sample size is always better. What this line of thinking led us to ignore was that in using the full sample, we would have lost some of the depth of detail that this longitudinal dataset offers and which qualitative coding is uniquely suited for exploiting.
We encourage other researchers working with large-scale qualitative datasets to consider limiting their sample size or the amount of data they use for their analyses. These decisions should be made strategically and with the study aims in mind (Roy, Zvonkovic, Goldberg, Sharp, & LaRossa, 2015). It may be possible to use the full dataset for some parts of the analysis and an analytic sample for others. We were able to use the full sample in our initial analysis that identified four transition types largely because of how the summary documents were structured. The timelines allowed us to easily catalogue all of the major childcare (and employment and housing) transitions for one family and the family profiles allowed us to quickly understand the most important details for these transitions and other attempted transitions.
The summary documents allowed us to quickly orient ourselves to the dataset and begin analysis. We were fortunate to not have to create these ourselves, but would encourage anyone working with a large dataset to expend the resources and time to create similar materials. There are many strategies for doing this and the exact format will be dictated by the study structure and aims, but the goal should be to produce short documents that provide an overall picture of the data available for each unit of analysis. It was also important that we were able to link the information in the timelines and family profiles back to the interviews and fieldnotes (through file names embedded in the summary documents).
Although summary documents are valuable resources, it is important to consider how their content and structure may affect analytic strategies and conclusions. Our decision to focus on childcare transitions may have been influenced by using the timelines—which inherently highlight transitions—to orient ourselves to the dataset. It is also likely that the information emphasized in the timelines (and to a lesser extent the more comprehensive family profiles) influenced our thinking about the areas of the families’ lives to consider. For example, because housing and health were included on the timelines it is likely that we gave them more attention (at least initially) than other aspects of the families’ lives not captured on the timelines (e.g., intimate relationships).
Finally, it is important to note that these analyses took place over several years. They reflect a “generative” scenario in which a young scholar shared a rich and promising dataset with two doctoral students, who then took the lead in conducting the analyses as they left graduate school and emerged as independent scholars. The depth of detail and volume of data found in large-scale qualitative datasets allows for multiple generations of family scholars to develop a long line of supported inquiry.
Using a large-scale qualitative dataset to explore an interesting research question offers unique challenges and rewards. The amount of data available can be daunting and may require producing short summary documents and selecting a smaller analytic sample. However, the depth of detail that this kind of dataset offers can be unparalleled. We found it particularly rewarding to be able to explore childcare transitions in a longitudinal dataset that allowed us to observe the transitions before, during, and after they happened. It is our hope that this chapter inspires and prepares qualitative researchers to conduct their own analysis of a large-scale dataset.
Adams, G., & Rohacek, M. H. (2010). Child care instability: Definitions, context, and policy implications. Washington, DC: The Urban Institute.
Chaudry, A. (2004). Putting children first: How low-wage working mothers manage child care. New York, NY: Russell Sage Foundation.
Hunt, G., Moloney, M., & Fazio, A. (2011). Embarking on large-scale qualitative research: Reaping the benefits of mixed methods in studying youth, clubs and drugs. Nordic Studies on Alcohol and Drugs, 28, 433–452.
LaRossa, R. (2005). Grounded theory methods and qualitative family research. Journal of Marriage and Family, 67, 837–857.
Lowe, E. D., Weisner, T. S., & Geis, S. (2003). Instability in child care: Ethnographic evidence from working poor families in the New Hope intervention. New York, NY: MDRC. Retrieved from www.mdrc.org/publication/instability-child-care
Maxwell, J. A., & Miller, B. A. (2008). Categorizing and connecting strategies in qualitative data analysis. In S. N. Hesse-Biber & P. Leavy (Eds.), Handbook of emergent methods (pp. 461–477). New York, NY: Guilford Press.
Pilarz, A. R., & Hill, H. D. (2014). Unstable and multiple child care arrangements and young children’s behavior. Early Childhood Research Quarterly, 29, 471–483.
Roy, K. M., & Burton, L. (2007). Mothering through recruitment: Kinscription of nonresidential fathers and father figures in low-income families. Family Relations, 56, 24–39.
Roy, K. M., Tubbs, C. Y., & Burton, L. M. (2004). Don’t have no time: Daily rhythms and the organization of time for low-income families. Family Relations, 53, 168–178.
Roy, K. M., Zvonkovic, A., Goldberg, A., Sharp, E., & LaRossa, R. (2015). Sampling richness and qualitative integrity: Challenges for research with families. Journal of Marriage and Family, 77, 243–260.
Speirs, K. E., Vesely, C. K., & Roy, K. (2015). Is stability always a good thing? Low-income mothers’ experiences with child care transitions. Children and Youth Services Review, 53, 147–156.
Tran, H., & Winsler, A. (2011). Teacher and center stability and school readiness among low-income, ethnically diverse children in subsidized, center-based child care. Children and Youth Services Review, 33, 2241–2252.
Van Horn, M. L., Ramey, S. L., Mulvihill, B. A., & Newell, W. Y. (2001). Reasons for child care choice and appraisal among low-income mothers. Child and Youth Care Forum, 30, 231–249.