Research Design
In order to investigate the role of childbearing in the lives of single mothers, we combined qualitative and quantitative methods. The mixed-method approach is increasingly being advocated in the social sciences (e.g. Bryman 1988; Giele and Elder 1998; Sale et al. 2002). Using different approaches, as well as different methods and data sets within each paradigm (methodological triangulation), allows us to formulate interpretations of social phenomena that are deeper and more valid. In our study, we employ qualitative methods—i.e., content analysis of semi-structured interviews—to identify the positive and the negative aspects of lone motherhood, and to reconstruct women’s perceptions of how giving birth while single has affected their lives. In the next step, we apply quantitative methods—i.e., panel econometric techniques—to assess the overall impact of having a child on single mothers’ happiness.
By integrating these two methodological approaches into a single research context we are able to offer a more comprehensive picture of single motherhood, which was not possible in previous studies that used a qualitative or a quantitative approach only. First, we are consistent with respect to the population under study. While previous qualitative research usually referred to never-married women, the quantitative research investigated all single mothers, including the divorced and widowed. We focus on lone mothers who were never married in both parts of our study, but in the quantitative part we also display results for previously partnered mothers. Second, we do not restrict our sample to teenage women from poor neighborhoods, as has often been the case in qualitative studies on this topic. Instead, we study women from various social backgrounds, which allows us to gain a wider perspective on the role of childbearing in lone mothers’ lives. Third, we do not restrict our analysis to narrowly defined or indirect indicators of well-being, but instead attempt to provide an overall subjective evaluation of the mothers’ happiness.
The advantages of our study lie not only in its mixed-methods design, but also in the characteristics of our quantitative component. In contrast to the available quantitative studies, which compared the subjective well-being of married and single mothers, we use longitudinal data to investigate how the arrival of a child changes the happiness of women with different marital and partnership statuses. Additionally, we apply econometric techniques that eliminate bias from the selection of women who are “innately unhappy” (due to childhood experiences or personality traits) into the group of lone mothers. Our quantitative analyses address the question of whether single women would be indeed happier if they had not given birth to a child, instead of the question of which group of mothers, single or married, are happier.
The Qualitative Study
Participants
Our qualitative data come from semi-structured face-to-face interviews which were conducted in 2011 within the research project “Family change and subjective well-being” (FAMWELL). The aim of the project was to explore new and currently rare family developments in Poland. In particular, the project seeks to investigate what life-course developments and circumstances lead individuals to arrive at certain family arrangements (e.g., lone motherhood), and to study how these developments affect individuals’ subjective well-being. Within this project, we conducted 35 interviews with women who experienced an extramarital birth. The interviews were conducted in cooperation with TNS OBOP research agency. The agency recruited the respondents using a snowball method: in several locations in Poland (three voivodeships; three towns or cities in each of them) the networks of the agency pollsters were used to snowball for women who were aged 25–39 and who had ever experienced an extramarital birth. Out of the 35 women recruited, 16 were cohabiting with their child’s father at the time of the interview, and they were excluded from the analyses. Another three women were raising their child with the child’s father for a prolonged period of time, and they were also dropped from the sample. This left us with a final sample of 16 women. None of these women was married before giving birth. In 12 cases, their relationship with the child’s father ended during the pregnancy. The remaining four women had separated from the child’s father at some point after the birth of the child (1–4 years). We decided to include them in the sample because all of them reported very serious problems in their relationships during the pregnancy; thus, even though the final termination of the relationship took place later, they said they felt like they were “single mothers” from the very beginning. Eight women in the sample were in a relationship with a new partner at the time of the interview, while the other eight were single. They all, however, experienced periods in which they were raising their child without any support from a partner during the early stages of the child’s life.
Our interviewees were 26–38 years old, and their main characteristics are presented in the table below. Importantly, we did not limit our sample to teenage mothers, nor did we select women from any particular city, neighborhood, or social group. The study was conducted in several locations in Poland; in large cities as well as in small towns. We looked for women from different social backgrounds and with different educational levels (Table 1).
Table 1 The structure of the sample for qualitative analyses The heterogeneity of the sample is apparent if we also consider the occupations of the respondents. We interviewed shop assistants, a cleaner, a hairdresser, an employee of a wholesale poultry vendor, a sales agent, an insurance agent, a social worker, an assistant in a law firm, and office workers.
Measurement Instruments
The interviews were semi-structured and problem-centered. The interview guideline was designed to reconstruct a history of how the respondent became a lone mother, and to explore how this event influenced her life and general level of happiness. Three thematic areas were covered in each interview. First, the respondents were asked to describe their life situations, with a special focus on their family and living arrangements. Next, questions designed to reconstruct a history of how the woman became a lone mother were asked. This second section always started with a question about how the respondent had imagined her family life as an adult when she was young. The respondents were asked to describe their desires and intentions related to family formation, and then to discuss the factors that, in their opinion, encouraged or discouraged the realization of these intentions. The respondents were prompted to consider various life events that might have been relevant (related to education, work, and relationships), and to discuss the roles played by other people (family, friends). In the third and final section of the interview, each woman was encouraged to imagine what her life would have been like if she had ended up in a more “traditional” family arrangement. In this section, each respondent was asked whether, on the whole, she would have been more or less happy if she had raised her child with the child’s father. This question was followed by questions on the respondent’s overall sense of life satisfaction and on the main sources of her feelings of happiness or distress.
All of the interviews were conducted following the above scenario, but the interviewers were allowed to adjust the wording of the questions to fit the interview flow and the specific characteristics of each respondent. The interviews were conducted by four experienced qualitative interviewers, who were women aged 22–32 (similar to the ages of the respondents, which allowed them to build a better rapport). All of the interviewers were instructed, coached, and supervised by the study coordinator.
Procedure and Data Analysis
A content analysis of the interviews was conducted to identify all of the positive and negative aspects of lone motherhood, as perceived by the respondents. The coding was performed by the coordinator of the qualitative study (a 35-year-old woman with a degree in psychology and several years of experience in collecting and analyzing qualitative data). We analyzed the data using a bottom-up coding procedure. NVivo 9 software was used to facilitate the process. First, we identified all of the passages in which any reference to childbearing and lone motherhood was made. This material was coded using the open coding procedure (Strauss and Corbin 1998). Next, the open codes were merged into categories. Nine categories emerged, which are presented in a table in the Appendix. In the next step, a matrix of all of the interviews and the categories was created. The matrix contained the key statements made by the respondents or short summaries of what they said during the interview in each of the categories. This matrix allowed us to conduct an efficient analysis of the content of each category, and made it easy to retrieve key quotations. The negative or positive aspects of lone motherhood, as revealed by the respondents, were identified for each category.
The Quantitative Study
In a second step, we turned to quantitative methods to estimate the general impact of giving birth to a child on the happiness of the unpartnered women. To this end, we used survey data from Social Diagnosis. Social Diagnosis is a panel multi-purpose survey designed to provide a regular assessment of the living conditions and the quality of life of the Polish population. To the best of our knowledge, it is the largest and most comprehensive panel survey carried out in Central and Eastern Europe that includes questions on happiness (Filer and Hanousek 2002). The individual-level data and survey documentation are available in the public domain (on the website www.diagnoza.com).
Social Diagnosis was conducted for the first time in 2000 on a random sample of 3,005 households. All of the household members aged 16 or above were supposed to be interviewed. They were interviewed again in 2003, and every 2 years thereafter. At each wave information on the respondents’ living conditions, family situations, education, labor market participation, health, and various aspects of subjective well-being (including general happiness) was collected.
Participants
Altogether, in all six waves, 65,282 face-to-face interviews were conducted (Czapiński and Panek 2011). For our analysis, we used data from the second and subsequent waves, as the question measuring happiness in the first wave was not comparable with the questions in the following waves. We selected women who entered the survey at ages 18–35; i.e., at childbearing and childrearing ages. This gave us a sample of 15,246 female observations. Around half of them (7,633) were mothers, among whom 538 were never married, 6,594 were married, and 501 were previously married (divorced or widowed). Cohabiting women were dropped from our sample, as we found only 87 such cases.
Measurement Instruments
In our study, we measured the self-rated general level of happiness, derived from a single-item question: “In general, would you say you are very happy, quite happy, somewhat happy, or not at all happy?”; with responses coded on a four-point scale. In the context of this study, this measure has the advantage of brevity. It was adapted from the World Value Survey, and a similar question is also included in other large cross-national or country-specific surveys. Single-item measurement is considered to be less reliable than multi-item scales, but an overall level of happiness is frequently measured with only one question, providing scores of satisfactory validity and reliability (e.g., Holder et al. 2010; Holder and Klassen 2010; Swinyard et al. 2001; Abdel-Khalek 2006).
Our main explanatory variable was created through an interaction of the fact of having a child with a woman’s marital status. Among our control variables, we included a set of observed person-specific characteristics, such as the respondent’s age, educational attainment (including participation in education), self-rated health, self-rated income level, and the age of the youngest child.
Procedure and Data Analysis
We modeled the i-th respondent’s self-rated happiness at any point in time t as a function of our key explanatory variables (the fact of having a child interacted with a woman’s marital status—child_mstat) at time t, a set of the observed individual-level characteristics measured in the survey at time t (obs_characteristics), as well as unobserved individual time-invariant traits ui. Additionally, respondent’s self-rated happiness was subject to random error εit, which may capture random, idiosyncratic influences, such as good weather on the day of the interview or an exceptionally good mood of the respondent. Hence, the model can be written in the following way:
$$happiness_{it} = \beta_{0} + \beta_{1} \times child\_mstat_{it} + \beta_{2} \times obs\_characteristics_{it} + u_{i} + \varepsilon_{it}$$
(1)
where the parameter β0 represents a constant, β1 reflects the effect of a child-partnership status on happiness and β2 shows the effect of individual-level characteristics (age, education, satisfaction with health, satisfaction with income, labor market situation of the mother and her partner) on happiness.
The most common approach to controlling for individual-specific unobserved characteristics with the panel data is to estimate fixed-effects models. Fixed-effects models are based on the variation of the respondent’s characteristics across time, and hence remove the potential bias resulting from the selection of “intrinsically (un)happy” individuals into the group of lone parents. In this paper, we employed two different fixed-effects estimators which were developed specifically for models with ordered dependent variables: namely, the FCF estimator proposed by Ferrer-i-Carbonell and Frijters (2004), and the “blow-up and cluster” (BUC) estimator recently developed by Baetschmann et al. (2011). The former is probably the best known tool used for estimating the fixed-effects ordered logit models, but it yields inconsistent estimates on panel data with a small number of waves (Baetschmann et al. 2011). Because the latter was shown to be less sensitive to the number of panel waves (Baetschmann et al. 2011), it might be better suited to our data with five waves. In addition to using these fixed-effects models, we also estimated a correlated random-effects ordered probit model, which draws on the approach proposed by Mundlak (1978), and, like the fixed-effects models, allows us to remove the selection bias. It decomposes the unobserved time-constant individual effect ui into a random effect, which is uncorrelated with the explanatory variables and the mean values of the time-varying regressors that are expected to be correlated with the individual random effects (Mundlak 1978). The estimates produced by this method are least sensitive to the number of panel waves, and are more robust to the incidental parameter problem (Greene and Hensher 2010).