Welcome to follow

The Journal of Chinese Sociology

2025年10月6日，The Journal of Chinese Sociology（《中國社會學學刊》）上線文章Exploring causal complexity in mate preferences in China: a reanalysis of survey experiment data using qualitative comparative analysis (QCA)（《探索中國擇偶偏好中的因果復雜性——使用QCA對調(diào)查實驗數(shù)據(jù)的再分析》）。

| 作者簡介

許琪

南京大學社會學院教授。

主要研究方向：家庭社會學、人口社會學、社會分層、定量研究方法。

Abstract

This paper employs two methods: a fixed effects model and a fuzzy-set qualitative comparative analysis (QCA) to analyze the survey experiment data on mate preference from Chinese General Social Survey in 2021. The study reveals that while the conclusions of both methods were consistent, QCA proves more advantageous in capturing the complexity in mate preferences. Specifically, the influence of mate selection conditions on mate selection decisions is not isolated but bound together, leading to configurational features in nature. In addition, acceptance and rejection decisions are affected by different causal mechanisms, and regression analysis cannot capture the asymmetry of this causal relationship. QCA and regression models are well-suited for analyzing different causal questions; it is only through the complementary use of both methods that a synergistic effect can be achieved.

Keywords

Mate preferences; Causal complexity; Survey experiment; Regression model; Qualitative comparative analysis

Introduction

Mate preference refers to the ideal conditions or criteria individuals consider when choosing a romantic or marital partner. It is shaped not only by personal tastes and desires but also by the constraints of family institutions, cultural values and social customs (Xu 2000). Studies have found that societies often display similar patterns in mate selection standards; men tend to prefer women who are younger, more attractive and gentler, while women are more inclined to favor men with higher socioeconomic status (Buss 1989). With societal development and socio-cultural change, mate preferences among both men and women transform to varying degrees and exhibit class-based heterogeneity. Thus, the study of mate preference offers a valuable lens for examining social transformation and understanding the role of marriage in the reproduction of social stratification (Buss et al. 2001; Li and Xu 2004; Zhou et al. 2023). At the same time, mate preference plays a significant part in the formation and stability of marriage. In recent years, marriage rates in China have been steadily declining, while divorce rates have sharply risen. Some studies suggest that the increasing tendency among young people to postpone or opt out of marriage is linked to elevated expectations of an ideal partner (Miao and Huang 2022). Whether mate preferences are fulfilled also appears to be a key factor in the harmony of marital relationships (Zheng and Fang 2019). Therefore, understanding the basic characteristics of mate preference is of great importance for grasping the ongoing transformation of marriage in contemporary China.

Existing research on mate preference largely adopts two main approaches. The first is to infer mate preferences indirectly through the analysis of marriage matching patterns (Schwartz 2013; Ma et al. 2019). The second is to directly observe or ask individuals about their mate selection criteria through personal ads, surveys and other methods (Buss 1989; Xu 2000). Both approaches have limitations however. In order to overcome these shortcomings, Zhou et al. (2023) innovatively applied the survey experiment method to measure mate preferences among Chinese people, pointing the way for future research. However, whether through traditional approaches or the more recent use of survey experiments, researchers have primarily relied on regression models for data analysis, such as linear regression or logistic regression. The inherent assumptions of linearity, additivity and symmetry in regression models limit their ability to capture the full complexity of mate preferences. This limitation manifests in two specific aspects.

First, individuals tend to make a comprehensive evaluation of a range of attributes for a potential partner before deciding to accept or reject one. Therefore, what primarily influences the decision-making process is this consideration of various attributes (of mate selection criteria) rather than any single criterion in isolation. In regression analysis, however, what researchers obtain is the estimated net effect of a specific criterion on the outcome after controlling for all other variables. This approach ignores the interactive relationships amongst mate selection criteria and does not align with how mate selection decisions are actually made in real life. Although researchers can include interaction terms in regression models to account for variable interdependence, such interactions are typically limited to two or three variables. High-dimensional interactions involving more than three variables are rarely used in practice, and even when applied, the results are often difficult to interpret clearly. This constitutes the first major limitation of using regression models to analyze mate preferences.

Second, regression models are symmetric causal models, meaning that the explanation for the occurrence of a dependent variable is assumed to be the same as that for its non-occurrence. For example, Zhou et al. (2023) used a logit model to examine the factors influencing whether a hypothetical spouse would be accepted. In such models, the factors influencing acceptance are treated as identical to those influencing rejection, and the effects are assumed to operate in the same way, (as expressed through a single regression equation). This is a typical approach in regression analysis and reflects an implicit assumption of causal symmetry. However, in many cases, causal models are asymmetric in practice. For instance, in mate selection, not rejecting someone does not necessarily mean accepting them, and the reasons for rejection are often different from those for acceptance. Regression models are unable to distinguish between these two logics, which constitutes the second major limitation of using regression models to analyze mate preferences.

Using regression models to analyze mate preferences, then, presents numerous challenges, many of which are difficult to overcome within the current framework of regression analysis. To address these challenges, this study employs qualitative comparative analysis (QCA). Unlike regression models, QCA allows for the analysis of complex effects stemming from different combinations of independent variables and can separately examine the conditions under which the dependent variable occurs or does not occur. Therefore, this method effectively compensates for the limitations of regression analysis.

This research has two main objectives: first, to explore the causal complexity of mate preference using QCA; and second, to introduce QCA as an analytical method to the academic community using mate preference as an example. Although QCA was proposed as early as 1987 by Charles C. Ragin (1987) and has undergone substantial development since then, it remains underutilized by sociologists in China, and related courses are rarely offered in Chinese universities. I believe that QCA offers distinctive advantages in analyzing causal complexity and, when used in conjunction with regression analysis, can produce a synergistic effect. The remainder of this paper is structured as follows: first, a review of existing studies on mate preferences is provided; next, followed by an introduction to the basic concepts of causal complexity and QCA. Then, both regression models and QCA are applied to analyze the survey experiment data on mate preferences embedded in the 2021 Chinese General Social Survey (CGSS 2021), aiming to replicate the findings of Zhou et al. (2023) and to further explore the complexity of mate preference. The paper concludes by summarizing the research findings and discussing the conditions under which regression models and QCA are most appropriately applied.

Mate preferences:

theoretical and empirical perspectives

Inferring mate preferences

indirectly through

mate selection behavior

Most existing studies on mate preferences follow two main research strategies. The first is to infer underlying preferences indirectly through observable mate selection behavior. For instance, many studies on marital matching have found that individuals tend to marry those with similar socioeconomic status and demographic characteristics, a pattern known as homogamy. This has led to the inference that people generally have a preference for similarity in mate selection (Qi and Niu 2012; Schwartz 2013). In addition, however, heterogamous marriages that cross class and identity boundaries are also common in human societies. To explain this, scholars have proposed exchange theory, which posits that individuals tend to use their own advantageous resources to obtain desirable traits or resources from their partners in return (Schwartz 2013; Ma et al. 2019). While such inferences are plausible, they also carry certain limitations.

One such limitation, with regard to homogamy, is that it may result from either a preference for similarity or a preference for competition. The theory of competitive mate preferences argues that individuals tend to pursue the most desirable partners in the marriage market, rather than those who are most similar to themselves. However, if everyone adopts this strategy, the outcome of market competition would still lead to homogamy (Schwartz 2013). Some studies suggest that individuals are more inclined toward similarity-based matching in terms of religion, values and interests, whereas competitive preferences dominate in the domain of socioeconomic status (Hitsch et al. 2010). Other scholars, however, argue that people are reluctant to risk failure by dating someone significantly more attractive or highly sought after, and thus prefer partners with similar socioeconomic status, suggesting that homogamy may prevail in this domain as well (Li and Xu 2004). In sum, there remains considerable academic debate over the explanatory power of similarity-based versus competitive mate preferences. The root of this debate lies in the fact that both types of preferences can lead to homogamous marriage outcomes, making it impossible to distinguish between them based solely on observed matching results.

Another limitation, regarding heterogamy, is that there is also considerable academic debate surrounding the theory of marital exchange. Proponents of this theory often cite examples such as race-status exchange (Gullickson 2006) and beauty-status exchange (Taylor and Glenn 1976) to support their arguments. In contrast, critics point out that even in interracial marriages, Black and White individuals tend to match with partners of similar social status, a pattern that the race-status exchange theory fails to explain (Rosenfeld 2005). Regarding beauty-status exchange, opponents argue that previous studies have focused only on the observation that physically attractive women are more likely to marry high-status men, while ignoring the fact that attractive women often possess high status themselves and that high-status men are often physically attractive. As a result, these studies have misinterpreted matching in terms of appearance and status as exchange (McClintock 2014). At present, the academic debate over the theory of marital exchange remains unresolved. Fundamentally, this controversy arises because existing studies continue to infer preferences from observed marriage matching data, and such inference is marked with uncertainty.

Finally, numerous studies have pointed out that mate selection is not only influenced by individual preferences but is also largely constrained by the opportunity structure composed of potential partners (Schwartz 2013). In terms of homogamy, for instance, interpersonal contact and interaction are prerequisites for forming intimate relationships and entering marriage. Structural factors such as homophily in social networks and residential segregation significantly increase the likelihood of interaction among individuals with similar characteristics, thereby raising the probability of homogamous marriages (Xie et al. 2015). As for heterogamy, such patterns of union formation may simply result from individuals being unable to realize their preferences for homogamy due to the constraints of a limited opportunity structure and thus are unrelated to any conscious marital exchange (Schwartz 2013). In short, due to the constraints of opportunity structure, actual mate selection behavior may not directly reflect individual mate preferences. Therefore, inferring mate preferences from observed behavior is subject to significant limitations.

Directly measuring

mate preference

through surveys

To overcome the limitations of the first research strategy, some scholars have argued that mate preference should be measured directly through surveys. This leads to the second research strategy concerning mate preferences. A particularly influential example in this regard is a cross-national comparative study on mate preference published by evolutionary psychologist David M. Buss (1989). In this study, Buss used structured questionnaires to survey more than 10,000 respondents from 37 countries about their views on mate selection criteria. The results revealed substantial gender differences. Women were more likely to prefer men who were well-educated, financially stable, ambitious and older than they were, whereas men tended to favor women who were younger than them and physically attractive with domestic skills (Buss 1989). In another study, Buss and his colleagues analyzed survey data on mate preferences collected in the United States from the 1940s to the 1990?s and found that gender differences in mate preferences remained remarkably stable over these five decades (Buss et al. 2001).

Because the survey results showed that gender differences in mate preferences are both cross-culturally consistent and temporally stable, Buss argued that such differences could only be explained from the perspective of human evolution (Buss 1989). However, social scientists, particularly economists and sociologists, have contended that gender differences in mate preferences should be understood through cultural and institutional lenses. Becker (1993) argued that both men and women aim to maximize their personal benefits through marriage. Since men have a comparative advantage in market labor and women have a comparative advantage in domestic labor, marital exchange and gendered division of labor can maximize the overall output of marriage. This leads to systematic differences in mate preferences between the sexes (Ibid).

According to Becker’s theory of the economics of the family, gender differences in labor market status are the root cause of differences in mate preferences. Therefore, as women’s labor force participation and income levels rise, the mate preferences of men and women should converge (Oppenheimer 1988). However, recent studies have not observed such a trend (Van Bavel et al. 2018). In response, sociologists argue from the perspective of gender role ideology. Gender role theory argues that long-standing disparities in access to social resources between men and women have become solidified into role expectations. Men are expected to fulfill the role of breadwinner, while women are expected to devote more effort to domestic labor and caregiving activities (Thompson and Walker 1995). Although women’s socioeconomic status has improved significantly in recent years, deeply entrenched gender role norms remain difficult to change, resulting in little shift in the standards that men and women use to evaluate ideal partners.

Simulating mate selection decisions

through survey experimentation

In sum, Scholars have accumulated a substantial body of theoretical and empirical research on mate preference through questionnaire surveys. However, this research strategy also has its limitations. First, mate preferences are a sensitive topic, and respondents may not report their true thoughts honestly. Instead, they might provide answers that are more in line with prevailing social norms, resulting in social desirability bias. Second, questionnaire surveys typically list several mate selection criteria and ask respondents to rate the importance of each one individually. Yet in real-life situations, people often make acceptance or rejection decisions only after comprehensively evaluating multiple attributes of a potential partner. By contrast, questionnaire surveys are limited in their ability to reproduce this complex decision-making process.

To address the limitations of the questionnaire survey approach, Zhou et al. (2023) introduced the survey experiment method to the study of mate preferences. They constructed hypothetical profiles based on six variables: education, income, housing ownership, family background, age and physical appearance. They then asked respondents to evaluate how desirable each hypothetical person would be as a marriage partner. Although decisions made in an experimental context may differ from those made in real-life situations, and the experiment itself was limited to only six mate selection conditions that excluded variables such as personal values, the simultaneous presentation of multiple characteristics in the survey experiment helps to partially mask certain sensitive attributes. This thereby reduces social desirability bias in respondents’ answers. On the other hand, since respondents were required to evaluate the desirability of each profile based on a combination of multiple characteristics, this design also facilitates a better approximation of the complex mate selection decision-making process.

The original intention behind the experimental design of Zhou et al. (2023), was to better analyze the combined influence of various mate selection conditions on mate selection outcomes. Zhou et al. (Ibid) noted in their article that mate preferences are a combinatory criterion; people evaluate potential partners based on the overall utility derived from preferences across multiple dimensions. However, in their actual analysis, Zhou et al. primarily focused on estimating the net effect of each specific condition on mate selection decisions. Although their use of random assignment in the experimental design successfully minimized correlations both among mate selection conditions and between these conditions and respondent characteristics, making it well-suited for net effect analysis, the profiles presented to respondents in the experiment incorporated all six characteristics simultaneously. This means that it is the configuration of these conditions, rather than any single condition, that truly exerts influence.

In addition, Zhou et al. (Ibid) used the results of the regression analysis to calculate the willingness to pay for each mate selection condition, thereby assigning a monetary value to each condition. The theoretical assumption here is that any shortcoming in a potential partner can be compensated for by an increase in income. However, in real-life situations, people may regard certain mate selection criteria as necessary conditions, the absence of which cannot be compensated for by other advantages. Therefore, not all mate selection conditions are substitutable, and distinguishing between necessary and non-necessary conditions can help us better understand the applicability of the theory of marital exchange.

Finally, individuals make the decisions to either accept or reject in mate selection, but regression analysis typically uses a single model to explain both types of decisions, thereby overlooking the asymmetric reasoning processes behind them. To address the limitations in Zhou et al.’s (2023) research and to better analyze the causal complexity of mate preferences, this paper will reanalyze the survey experiment data on mate preferences using QCA. Given that the Chinese sociologists have limited familiarity with QCA, the following section will first provide a brief introduction to the concept of causal complexity and the basic logic of QCA.

Causal complexity and QCA

Causal complexity

Causal analysis has always been a central goal of scientific research, yet the path to uncovering causality is far from straightforward. In quantitative research, scholars primarily focus on estimating the average causal effect of an independent variable X on a dependent variable Y. However, the estimation of average causal effects is often subject to various sources of confounding influences. Therefore, for quantitative research, the complexity of causal analysis lies in how to eliminate all confounding factors in order to obtain the net effect of X on Y.

Unlike quantitative scholars, who strive to estimate the causal effect of X on Y, qualitative researchers are more concerned with identifying the cause of an effect. For instance, they often select multiple cases in which the outcome has already occurred and then explore in depth how that outcome came about. In this process, qualitative researchers pay particular attention to two distinct types of causal conditions: necessary conditions and sufficient conditions. A necessary condition implies that without it, the outcome cannot occur; a sufficient condition implies that with it, the outcome must occur (Goertz and Mahoney 2012).

In social science research, outcomes are rarely caused by a single factor. Therefore, when identifying sufficient conditions, qualitative researchers tend to focus more on combinations of multiple causes. For instance, factors A, B and C may all influence the outcome, but none of them alone is sufficient to produce the outcome; only when A, B and C appear together does the outcome occur. Such a combination of causal conditions is referred to as a configuration. In addition, for many social phenomena, there is often more than one causal path leading to the same outcome. For example, the co-occurrence of A, B and C may lead to the outcome, and so may the co-occurrence of B, C and D. In this case, both ABC and BCD configurations serve as sufficient conditions for the outcome. This phenomenon, where multiple distinct causal paths lead to the same outcome, is known as equifinality. Finally, the complexity of causal analysis also lies in the fact that the causes of the presence of an outcome are often different from the causes of its absence. This is referred to as asymmetry in causal relationships.

The basic principles of QCA

In sum, while scholars engaged in quantitative research have developed various scientifically rigorous tools for causal analysis, they have paid relatively little attention to certain issues. These include the distinction between sufficient and necessary conditions, the configurational features of sufficient causes, and the equifinality and asymmetry of causal relationships. In contrast, qualitative researchers have developed a more nuanced understanding of causal complexity, but they have long lacked a universal, transparent and methodologically rigorous approach for analyzing such complexity. In response to the limitations of both research traditions, Charles C. Ragin (1987) proposed a method that combines the methodological rigor of quantitative research with the substantive depth of qualitative inquiry and named it as qualitative comparative analysis (QCA).

QCA is an analytical technique grounded in logic and set theory. In practice, researchers must first transform variables into sets and determine each case’s membership in these sets. For example, “owning property” constitutes a set, and each respondent can be categorized as either belonging to or not belonging to this set based on their housing status. Once all sets are defined, set-theoretic methods can be used to determine whether a subset or superset relationship exists between them. If all members of Set A are also members of Set B, then Set A is a subset of Set B, or equivalently, Set B is a superset of Set A. According to Ragin (Ibid), these subset and superset relationships are key to identifying sufficient and necessary conditions.

For example, in the context of mate selection, if all those who are accepted (Y) possess property (A), A is a superset of Y, and “having property” (A) can be regarded as a necessary condition for being accepted (Y). Conversely, if all individuals with property (A) are accepted (Y), A is a subset of Y, and “having property” (A) constitutes a sufficient condition for being accepted (Y). In social science research, sufficient conditions are rarely constituted by a single factor; rather, they typically manifest as combinations of multiple conditions. For instance, in mate selection, having property (A) alone is usually insufficient to ensure acceptance (Y), but the combination of having property (A), high educational attainment (B) and a perceived attractive appearance (C), together, may constitute a sufficient condition for acceptance (Y). In this case, the combination ABC is a sufficient condition for Y; this combination is what is referred to as a configuration and can be derived by calculating the intersection of sets A, B and C, respectively. In practice, there may be more than one condition combination that leads to the same outcome. For example, the combination of high education (B), high income (D) and a perceived attractive appearance (C), may also constitute a sufficient condition for acceptance (Y). In this case, there are two causal paths to the same outcome Y: one is ABC, the other is BCD. This exemplifies the equifinality discussed earlier.

The above results can be represented using Expression 1:

It is important to note that the equal sign, plus sign, and multiplication sign in Expression 1 differ from their conventional mathematical meanings. Specifically, the equal sign in Expression 1 denotes a subset relationship in set theory, that is, the set on the right-hand side is a subset of the set on the left-hand side. Similarly, the multiplication and addition signs do not refer to arithmetic operations but instead indicate a set-theoretic intersection and union operations, or logical “and” and “or” operators, respectively. Solving expressions like Expression 1 is a key step in QCA analysis. As such, expressions clearly display all causal pathways that lead to the occurrence of Y and the constituent elements along each pathway (Du and Jia 2017). Researchers can also use the same approach to analyze cases where Y does not occur, thereby examining the asymmetry of causal relationships.

The analytical method described above is known as crisp-set QCA, which is primarily suited for situations where set membership is clearly defined. However, in some cases, the membership of a case in a given set is not so clear-cut. In such situations, fuzzy-set QCA becomes necessary. While the analytical logic of fuzzy-set QCA is similar to that of crisp-set QCA, it differs in two particular aspects.

First, before applying fuzzy-set QCA, it is necessary to calibrate the variables that convert them into membership scores in fuzzy sets. A membership score is a value ranging from 0 to 1, with a value closer to 1 indicating a higher degree of membership in the given set. Existing research typically adopts two main approaches to calibration. The first approach is theory-driven, where researchers determine the values of the variable at three key points (i.e. full membership, full non-membership, and the crossover point). They then use these values to construct a logistic distribution function, which yields the membership scores for each case in the fuzzy set (Ragin 2008). The second approach is that researchers rank the cases based on the variable and then standardize the rank values to scores between 0 and 1 (Longest and Vaisey 2008). Each method has its advantages and limitations in the context of fuzzy-set calibration. The first method has higher validity and is easier to interpret, but it requires strong theoretical justification and may be more subjective in the absence of clear criteria. The second method is more widely applicable across contexts and does not require additional parameters, but the resulting membership scores are often less interpretable from a theoretical perspective.

Furthermore, in fuzzy-set QCA, the operations for determining subset, superset, intersection and union, differ mathematically from those in crisp-set QCA. Specifically, in fuzzy-set QCA, if X is a subset of Y, then it must be the case that Xi?≤?Yi for all cases. Conversely, if X is a superset of Y, then Xi?≥?Yi. The intersection of X and Y is defined as the minimum of the two values for each case, that is, min(Xi, Yi); the union is defined as the maximum, that is, max(Xi, Yi). Based on these formulas, Ragin (2008) proposed the method to assess necessary and sufficient conditions in fuzzy sets. For example, to determine necessity, Ragin introduced the consistency index, which is calculated as: consistency?=?∑min(Xi, Yi)/∑Yi. This index ranges from 0 to 1, with values closer to 1 indicating a stronger fit with the criteria for necessity. In addition to consistency, Ragin also proposed the concept of coverage, which is used to assess the relative importance of different necessary conditions. The formula for coverage put forth by Ragin (2008) is: coverage?=?∑min(Xi, Yi)/∑Xi. This index also ranges from 0 to 1, with higher values indicating greater importance of X.

Data and variables

Data

This study is based on the 2021 data from the Chinese General Social Survey (CGSS 2021). The CGSS is a large-scale, nationally representative and comprehensive survey designed and implemented by the National Survey Research Center at Renmin University in Beijing. Launched in 2003, the CGSS has conducted multiple cross-sectional surveys across the country, each with a sample size of approximately 10,000 respondents. This study uses the most recent data collected in 2021.

The CGSS 2021 includes a survey experiment module designed to measure mate preferences that adopts the same design as that used in the study by Zhou et al. (2023). Analyzing this module, therefore, allows us to replicate and extend their research. Following Zhou et al. (2023), this study retains only respondents aged 50 and below in the CGSS 2021 dataset. After excluding cases with missing values, the final sample size is 3,470 respondents.

Research design and variables

The survey experiment on mate preferences in CGSS2021 was conducted in two steps. First, three hypothetical profiles were generated based on six predetermined indicators: age, income, family background, housing ownership, education and (perceived) physical appearance. Age was generated as a random integer based on the respondent’s gender. For male respondents, the age difference between the respondent and the hypothetical individual ranged from ? 15 to?+?5 years (i.e. 15 years younger to 5 years older). For female respondents, the range was ? 5 to?+?15 years (i.e. 5 years younger to 15 years older). The income of the hypothetical individuals included nine values: 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.5, 2 and 3 times the respondent’s own income. Family background was classified into two categories: parents from rural areas and parents from urban areas. Housing ownership included two categories: no property and owning property. Education was categorized into high school, undergraduate and graduate levels. Physical appearance was divided into three levels: somewhat unattractive, average, and relatively attractive.

Moreover, the résumés of the three hypothetical individuals were presented to respondents, who were asked to evaluate, one by one, how desirable each person would be as a potential marriage partner. The rating ranged from 1 to 7, with higher scores indicating greater satisfaction. Based on the experimental design described above, the dependent variable in this study is the score assigned by each respondent to each hypothetical individual, while the independent variables are the values of the six characteristics assigned to the hypothetical individuals. Given the clear gender differences in mate preferences, the analysis is conducted separately by gender. In addition, following the approach of Zhou et al. (2023), educational attainment is used as a proxy for social class to explore class-based heterogeneity in mate preferences.

Analytical results

Results of regression analysis

The data analysis in this paper will be conducted in two steps. In the first step, an individual-level fixed effects model is employed to estimate the average causal effects on respondents’ ratings of the six indicators detailed above. The objective of this step is to replicate the findings of Zhou et al. (2023) and to compare them with the results of the QCA analysis presented later. It should be noted that the survey experiment adopted a randomized design to control for all individual-level confounding factors. However, to further address potential threats to internal validity, this paper applies a fixed effects model to statistically control for individual characteristics. In addition, using a fixed-effects model also helps to mitigate the influence of within-individual autocorrelation on statistical inference.

Table 1 presents the analysis of gender differences in mate preferences using the fixed effects model. The results show that the effect of the virtual individual's age on respondents follows an inverted U-shaped quadratic curve for both men and women. For female respondents, the peak of the curve is at 2.3 years, indicating that the optimal partner age is 2.3 years older than the respondent. For male respondents, the peak is at ?2.4 years, meaning that the optimal partner age is 2.4 years younger than the respondent. Taken together, these findings suggest that men and women share a similar perception of the ideal age gap in a partner, both favoring a pairing where the man is approximately two years older than the woman.

In addition, the socioeconomic status of virtual individuals has a significant effect on the ratings given by both male and female respondents. Specifically, regardless of gender, respondents showed a stronger preference for partners with higher income, urban parental background, property ownership and higher educational attainment. However, the impact of these socioeconomic characteristics is more pronounced among women. Statistical tests comparing the regression coefficients by gender show that the effects of income, property ownership and education are significantly greater for women than for men. This indicates that women place more importance on the socioeconomic status of a potential male partner in mate selection, which is largely consistent with the findings of Zhou et al. (2023).

Finally, Table 1 also shows that both men and women attach great importance to the appearance of potential partners in the mate selection process. However, compared to women, (perceived) physical appearance has a stronger effect on the ratings of male participants, and this gender difference is statistically significant. This finding is also consistent with the conclusions of Zhou et al. (2023).

Next, we examine the class-based heterogeneity in mate preferences by educational attainment for both men and women. As shown in Fig. 1, regardless of gender, respondents with higher education tend to have higher expectations on the socioeconomic status and physical appearance of their potential partners. This difference, however, is more pronounced among men. Similar findings were also reported by Zhou et al. (2023), and readers may refer to their study for further details.

Results of the QCA analysis

The above section has reproduced the research of Zhou and colleagues using a regression model. This section will reanalyze the same dataset using fuzzy-set QCA to better explore the causal complexity of mate preferences.

Calibration of variables

This study employs the fuzzy package in Stata to conduct fuzzy-set QCA analysis. The fuzzy package includes a dedicated command for variable calibration, called “setgen”, which offers several built-in functions for this purpose. Following the recommendation of Longest and Vaisey (2008), this paper uses the “()” function for calibration. This function first ranks the values of a variable and then standardizes the ordinal values into a score between 0 and 1, which represents the membership score of the variable in the fuzzy set. Compared to other calibration methods, this approach does not require manual specification of parameter values, making it more suitable for exploratory research.

In practice, the respondent’s rating, as well as the attributes of virtual individuals, can be directly calibrated using the “stdrank()” function based on their original values, including income, family background, housing ownership, educational attainment and physical appearance. However, the effect of a virtual individual’s age on the respondent’s rating follows a quadratic curve. Therefore, a preprocessing step is required before calibration. As discussed earlier, the optimal age combination is when the male is two years older than the female. Based on this standard, we calculate the deviation of each virtual individual's age from this optimal point. We then use the “stdrank()” function to calibrate this age difference, thereby obtaining the membership score representing how closely the virtual individual’s age aligns with the respondents.

Analysis of

individual necessary conditions

The first step in QCA analysis is often to identify necessary conditions for the outcome. Following Ragin’s (2008) recommendation, this study adopts a consistency threshold of 0.9 to determine whether a condition qualifies as necessary. As shown in Tables 2 and 3, regardless of gender or whether the analysis focuses on acceptance or rejection decisions, none of the conditions has a consistency score exceeding 0.9. Therefore, no single condition qualifies as a necessary condition for either acceptance or rejection. Some scholars argue that the primary purpose of analyzing necessary conditions in QCA is to filter out conditions suitable for configurational analysis (Schneider and Wagemann 2012). From this perspective, since none of the six conditions pass the necessity test, all can be retained as antecedent conditions for subsequent configurational analysis.

Configurational analysis

of conditions for acceptance

Next, we conduct a configurational analysis of all six conditions to examine whether specific combinations of conditions constitute sufficient circumstances for accepting or rejecting a hypothetical individual. Following the recommendation by Ragin (2008), this paper uses a consistency score threshold of 0.8 to determine sufficiency. In addition, 0.8 is also the default threshold used in the fuzzy command of the Stata package. Longest and Vaisey (2008) suggest that when the number of cases is relatively large, it is advisable to statistically test whether the consistency score significantly exceeds 0.8. Given the large sample size in this study, we adopt their suggestion and set the threshold at a consistency score that is significantly greater than 0.8 at the 0.05 significance level. Furthermore, the fuzzy command by default only analyzes configurations with more than one case. In this study, due to the large sample size and the random assignment of mate selection conditions in the survey experiment design, the number of cases for each condition configuration substantially exceeds the minimum threshold of one required for analysis.

This paper follows the presentation format for condition configurations as proposed by Ragin (2008): “●” indicates the presence of a condition, “○” indicates its absence, and a blank indicates that the condition is irrelevant. Table 4 presents the results of the fuzzy-set QCA analysis on the condition configurations for accepting a hypothetical partner, disaggregated by gender. It shows that women tend to make decisions of acceptance under two types of condition combinations: (1) similar age, high income, homeownership and (perceived) attractive appearance; and (2) similar age, urban parental background, homeownership, higher education level and (perceived) attractive appearance. Between these two configurations, Configuration 1 has higher coverage and unique coverage, indicating that it has stronger explanatory power. In other words, the combination of conditions in Configuration 1 is more widely accepted by women. Comparing Configuration 1 and Configuration 2, we find that they share three common conditions: similar age, homeownership and (perceived) attractive appearance. Although these three conditions did not pass the necessity test for single conditions mentioned earlier, their simultaneous presence in all configurations suggests that they are more important than other conditions in influencing preferred mate selection for women.

The analysis results for men show that six condition configurations are associated with acceptance, significantly more than the two configurations identified for women. From this perspective, men have a broader acceptance domain and exhibit greater heterogeneity in mate preferences. Synthesizing these six configurations, we find that similar age and (perceived) attractive appearance appear together in five of them, while the other four socioeconomic conditions appear together in no more than three. This indicates that, compared to socioeconomic status, men place greater emphasis on the age and appearance of women in mate selection. This interpretation is further supported by another observation. In Configurations 1, 2 and 3, respectively, when both similar age and (perceived) attractive appearance are present, the addition of just one or two socioeconomic conditions is sufficient for men to accept a preferred mate. However, when either similar age or (perceived) attractive appearance is not met (as in Configurations 5 and 6), men decide a preference only when all four socioeconomic conditions are simultaneously satisfied. This suggests that men have relatively strict expectations regarding a partner’s age and appearance. If one of these is lacking, they tend to place greater weight on other conditions. Furthermore, a comparison of male and female mate preferences reveals that while women consistently prefer men with high socioeconomic status, men do not necessarily prefer women with high socioeconomic status. For instance, Configuration 4 shows that women whose parents live in rural areas and who do not own property may still be favored by some men, likely due to the influence of traditional gender norms or mate selection gradient effects.

Next, we analyze the class-based heterogeneity in acceptance decisions by educational attainment. Table 5 presents the results for women. It shows that women without higher education exhibit three valid configurations meeting the sufficiency and consistency criteria in the QCA analysis. Among them, Configuration 1 and 3 are identical to those identified in Table 4 for all women, while Configuration 2 is a new solution. The most notable feature of this configuration is that having parents living in urban areas negatively influences acceptance decisions among women without higher education. This may be because many of them come from rural backgrounds and therefore exhibit a preference for homogamy in terms of household registration status. For women with higher education, only one configuration meets the criteria; they tend to prefer men who are similar in age, have high income, own property, possess a high level of education and have a (perceived) attractive appearance. This suggests that highly educated women place particularly high demands on men’s socioeconomic status, requiring all three conditions to be simultaneously fulfilled, including high income, higher education and property ownership.

Table 6 presents the results for men. It shows that regardless of educational attainment, men tend to prefer women who are similar in age and have a (perceived) attractive appearance. However, the effect of the potential partner’s socioeconomic characteristics varies according to the man’s level of education. For men with higher education, the influence of a partner’s socioeconomic status is positive. In contrast, for men without higher education, the impact of a partner’s socioeconomic status is less clear. In some configurations, the effect is positive, while in others, it is negative. This finding suggests that mate preferences among lower-class men are shaped by two competing tendencies. On one hand, influenced by competitive mate selection preferences, some men from a lower socio-economic class express a preference for women from higher socioeconomic backgrounds. On the other hand, due to more traditional gender norms and a tendency toward assortative mating, they also show a preference for women from lower socioeconomic backgrounds. As the regression analysis above demonstrates, the socioeconomic status of the virtual profile had a limited impact on mate selection decisions among men with lower educational attainment. This may be because regression analysis, in estimating average causal effects, combines these two divergent preferences. In contrast, QCA can distinguish and present these preferences through different configurational pathways, thereby enriching our understanding of mate preferences among men from a lower socio-economic class.

Configurational analysis

of conditions for rejection

This section analyzes the configurational conditions under which men and women make decisions to reject a potential mate. As shown in Table 7, there are eight configurations associated with the decision to reject by women, whereas only two configurations were found for their acceptance decisions in Table 4. This suggests that while the reasons for acceptance tend to converge for women, the reasons for rejection are more diverse.

Overall, the insufficient socioeconomic status of the male partner is the main reason for female rejection. Across the eight configurations, lacking property, low income and low education each appeared five times. However, the effect of socioeconomic status on rejection also depends on the potential partner’s age and appearance. If the partner has both an age and appearance that deviate significantly from respondents’ preferences, even one deficiency in either education or income is sufficient to trigger rejection (Configurations 1 and 2). If the partner fails to meet only one of the two criteria (i.e. age or appearance), then at least two deficiencies in socioeconomic traits are required for rejection (Configurations 3 to 7). If the partner’s age and appearance are both acceptable, then rejection occurs only when all four socioeconomic conditions are simultaneously absent (Configuration 8).

In contrast, only three configurations are associated with rejection decisions made by male respondents. None of these includes the condition of a (perceived) attractive appearance,, indicating that for men, a partner’s appearance has greater weight in making rejection decisions. Moreover, the conditions for rejection by males are more stringent. The potential partner must show deficiencies in at least four of the six conditions to trigger rejection, whereas female respondents may reject male partners based on three deficiencies. Lastly, having parents who live in urban areas may lead to rejection by some men (Configuration 3), but this is not the case for women. This may reflect the perception among some lower-status men that they are unlikely to be accepted by women from urban backgrounds.

Next, we analyze the class-based heterogeneity in rejection decisions from both male and female respondents by educational attainment. Table 8 presents the analysis results for women. It shows that highly educated women are associated with ten configurational conditions for rejection decisions, which is more than those found for women without higher education. Moreover, highly educated women may reject a potential partner when only two deficiencies are present, whereas women without higher education tend to reject only when the potential partner has at least four deficiencies. This suggests that women with higher socioeconomic status reject for a wider range of reasons and have lower thresholds for rejection, which is likely related to higher mate selection standards.

Table 9 presents the analysis results of rejection decisions among men by educational attainment. It shows that men without higher education are associated with only one configurational condition for rejection, and this configuration includes all six conditions simultaneously. This indicates that less-educated men rarely reject, and when they do, the criteria for rejection are extremely stringent, likely due to their disadvantaged position in the marriage market. In contrast, highly educated men are associated with a greater number of rejection configurations, each with relatively less demanding conditions. However, compared to highly educated women, the number of configurations linked to rejection decisions among highly educated men remains comparatively less, and the criteria for rejection are still somewhat stricter.

Robustness check

This study conducts robustness checks of the above analysis results from four aspects. First, in addition to the calibration method used earlier, this study also attempts to calibrate variables by specifying full membership, full non-membership, and crossover points. Specifically, the maximum and minimum values of each variable are used as the thresholds for full membership and full non-membership, respectively. While the median of the distribution is chosen as the crossover point. The findings show that the results based on this calibration approach differ from those above only in the solutions of a few configurations, whereas the main conclusions remain robust.

This study further adopts alternative criteria for determining sufficient conditions. First, following Ragin’s (2008) recommendation, the study used a consistency score greater than 0.8 as the threshold, without requiring statistical significance at the 0.05 level. Second, a stricter threshold of 0.85 was employed. The results show that using a criterion of a consistency score above 0.8 leads to the inclusion of more configurational solutions and increases the complexity of the results. The core conclusions, however, remain as robust as reported earlier. Meanwhile, adopting the 0.85 threshold yields results nearly identical to those obtained using the significance-based criterion, thus confirming the robustness of the above findings.

Moreover, the CGSS mate preferences experiment provided three scenarios for each respondent. In applying fuzzy-set QCA, this paper considered the three responses from each individual as three separate cases, which risks overlooking the potential influence of within-group autocorrelation. Therefore, two supplementary analyses were conducted. First, we used the cluster() option in the fuzzy command to account for within-group autocorrelation. The results showed that applying this option had almost no effect on the conclusions. Second, we conducted an analysis using only the first response of each respondent. Although this significantly reduced the sample size, the results remained robust. Thus, the influence of autocorrelation on the analysis is minimal.

Lastly, following the approach of Zhou et al. (2023), this paper also used respondents’ household registration (hukou) status as an alternative indicator of social class. The analysis showed that the results based on hukou status were entirely consistent with those based on educational attainment. Therefore, the findings regarding class heterogeneity in this study are supported across alternative measures of social class.

Conclusion and discussion

This paper uses both the fixed effects model and fuzzy-set QCA to analyze survey experiment data on mate preferences from the CGSS 2021. The study finds that while the conclusions of both methods are largely consistent, QCA proves to be more effective in capturing the causal complexity of mate preferences. This complexity mainly manifests in the following two aspects.

First, the influence of mate selection conditions on mate selection decisions is not isolated but interdependent, thereby exhibiting configurational features in two ways. On the one hand, individuals make decisions to accept or reject a mate only when multiple conditions are simultaneously satisfied or unsatisfied. Such complexity is difficult to uncover through regression analysis. On the other hand, the QCA results further reveal that the same condition may have different causal directions across different configurations. For example, the analysis of men from a lower socio-economic class reveals two distinct preferences within this group; some prefer women with higher socioeconomic status, while others prefer women with lower socioeconomic status. Regression analysis, by averaging causal effects, blends these divergent preferences and therefore fails to capture the heterogeneity in mate preferences among said male respondents.

Second, different mate selection conditions are governed by distinct causal mechanisms in acceptance and rejection decisions. Regression analysis has difficulty uncovering this asymmetry in causal relationships. Specifically, the QCA results show that the conditions under which women accept a potential partner are relatively clear-cut, while the conditions for rejection are more varied. Moreover, women tend to hold much higher standards for accepting a partner than for rejecting one, especially among those with higher socioeconomic status. In contrast, men demonstrate broader acceptance thresholds but stricter standards for rejection. For men with lower socioeconomic status, who are generally disadvantaged in the marriage market, rejection decisions are made only when potential partners fail to meet particularly stringent conditions.

Thus, QCA enables the analysis of complex effects arising from various combinations of mate selection conditions on mate selection decisions. It also allows for the separate examination of the causal mechanisms behind acceptance and rejection. This might suggest that QCA is better suited for capturing configurational causal complexity compared to regression analysis. Indeed, since Ragin introduced the concept of QCA, some scholars have considered it the start of a potential “Ragin revolution” in the social sciences (Vaisey 2009). However, this paper argues that while QCA offers unique advantages in analyzing causal complexity, it may be premature to claim that QCA will replace regression analysis. Overall, QCA and regression analysis are designed for different types of causal questions and are based on distinct underlying assumptions about causality. Therefore, rather than being substitutes, the two methods should be seen as analytically complementary. This complementarity manifests in two key ways.

The primary goal of regression analysis is to estimate the average causal effect of independent variables on the dependent variable, making it more suitable for analyzing the results of a given cause. In contrast, the primary goal of QCA is to explore the configurational conditions under which an outcome occurs or does not occur, making it more appropriate for analyzing the causes of a given result. For example, if we want to understand how much one’s attractiveness in the marriage market would change after a certain trait is altered, then regression models are more appropriate. However, if we aim to find out which combinations of traits are more likely to lead to acceptance in mate selection, QCA is the better choice. Since analyzing the result of a cause and the cause of a result are two different aspects of causal analysis, there is no hierarchy between them. Therefore, researchers must first clarify their research question before selecting the most appropriate analytical tool.

Following this, the mathematical foundation of regression analysis lies in probability theory and statistics in the social sciences, which implies an assumption of probabilistic causality. Specifically, regression models always include an error term to capture random factors that cannot be observed but in fact influence the dependent variable. Moreover, regression models do not assume that the effect of independent variables on the dependent variable is constant; rather, the coefficients obtained through regression analysis are merely estimates of the average causal effect. In contrast, QCA is based on logic and set theory, two mathematical tools that do not allow for exceptions in their calculations, thereby implying an assumption of deterministic causality. Although Ragin recognized that assuming no exceptions is too stringent for social science research, in practice, neither necessary nor sufficient conditions are strictly judged by a consistency score of 1. Nevertheless, when researchers use the concepts of sufficient and necessary conditions, they are effectively discussing some form of deterministic causality. In the author's view, social phenomena involve both deterministic and probabilistic elements. For example, in mate selection, certain combinations of conditions may indeed create strong influences on partner choice, yet people’s mate selection decisions are not entirely mechanical or predictable. Therefore, for the same social phenomenon, both regression analysis and QCA have their respective uses. The two methods used as analytical approaches can complement each other but cannot substitute for one another.

Although regression analysis and QCA serve different purposes and are suitable for different research contexts, Chinese sociologists are noticeably less familiar with and less likely to use QCA compared to regression analysis. Only a limited number of studies employing QCA have been published in leading sociology journals such as Sociological Studies (Huang and Gui 2009; Huang et al. 2015; Wang 2023). In contrast, QCA has seen broader application in disciplines such as management studies and communication studies (e.g. Du and Jia 2017; Sun and Zhang 2022). In fact, Ragin, who developed QCA, received his PhD in sociology from the University of North Carolina and has long been engaged in sociological research. Therefore, QCA ought to be more widely recognized and utilized within the field of sociology. One of the goals of this paper is to introduce QCA to Chinese domestic sociologists through the example of mate preference selection. It is hoped that this article will serve as a modest starting point and encourage more high-quality sociological research using QCA in China.

免費閱讀并下載全文

引用本文

Xu, Q. Exploring causal complexity in mate preferences in China: a reanalysis of survey experiment data using qualitative comparative analysis (QCA). J. Chin. Sociol. 12, 18 (2025). https://doi.org/10.1186/s40711-025-00245-z

https://journalofchinesesociology.springeropen.com/articles/10.1186/s40711-025-00245-z

文章僅為作者觀點，不代表本刊立場

以上就是本期JCS推文的內(nèi)容啦！

定期查收講座、征文信息/趣文推薦/熱點追蹤/主題漫談

學術(shù)路上