Drive to Marry and Social Prescription in Chinese Online Daters

Normative Pressures

We note that Blakemore et al.’s conceptualization and measurement of DTM emphasized personal over social prescription. In this context, prescriptive support is defined as a sense of obligation to remain in a relationship fueled by either personal or social reasons (see, e.g., Cox, Wexler, Rusbult, & Gaines, 1997). A personal reason (“personal prescription") involves internalized beliefs that advocate remaining in a relationship, whereas “social prescription” refers to believing that significant members (or “referents”) of one's social network support persisting in a relationship, for either moral or pragmatic reasons. Thus, the notion of social prescription coincides with the same role as subjective norms included in the classic volitional decision-making model of the Theory of Reasoned Action (or Planned Behavior) (Ajzen, 1991; Ajzen & Fishbein, 2005; Fishbein & Ajzen, 1975; Sniehotta, 2009; for a review see: Armitage & Conner, 2001; Sheppard, Hartwick, & Warshaw, 1988). Here, subjective norms are seen as one of the factors that determine one’s intention to perform a particular behavior.

Cox et al. (1997) found support for social prescription in commitment formation but not for personal prescription, i.e., the belief that one "ought to persist" influences feelings of commitment in ways extending beyond "wanting to persist," "feeling bound to persist," or "having no choice but to persist" (p. 87). These authors concluded that many of the previous studies on relationship commitment have been shortsighted due to neglecting the role of prescriptive support. It seems reasonable to expect that the same also applies to the DTM construct. Therefore, potential biases related to normative factors should be considered, including pressures from sources like friends, family, colleagues, and society at large.

Of course, the prevalence and impact of such pressures is likely to vary across communities or cultures. For example, past research indicates that attitudes toward commitment in a relationship differ between Chinese and American societies (Tzeng, 1993). Specifically, Hsu (1981) observed that for the Chinese, the terms “romantic relationship” or “dating relationship” imply the elements of necessary seriousness and long-term commitment, so romantic relationships are often perceived as one step before marriage. On the other hand, a committed relationship in American culture tends to be grounded in strong emotional experiences (e.g., Dion & Dion, 1988). Therefore, commitment typically has a more pragmatic function in the Chinese culture, which often reflects the obligation or responsibility of a person that is involved in a relationship from a more collectivist, versus Western-based individualistic, orientation (Bagozzi, Wong, & Yi, 1999; Hatfield & Rapson, 2002; Hsu, 1985; Markus & Kitayama, 1991; Wang, 2004). Accordingly, it might be expected that a serious desire, anticipation, and intent to get married would be markedly evidenced in a sample of Chinese singles who are currently using a matchmaking service to find a spouse. Indeed, Lange et al. (2015) recently validated Desire to Marry as an important variable that single men and women in China use to evaluate the suitability of romantic prospects.

The Present Research

Obviously it would be ideal to conduct a formal cross-cultural study on DTM directly comparing Eastern and Western samples. However, at the present time we have been only successful in gaining access to a verified sample of active online daters residing in China. The present research thus forms a first step that focuses on the DTM’s psychometric properties and biases in contemporary Chinese samples, thereby cross-validating and extending Blakemore et al.’s (2005) conceptualization and measurement of the construct. Two sets of questions will be addressed. First, although the DTM items were found to correlate sufficiently to form factors does not change the fact that the scores it produces are ordinal indices at best.

As described in more detail in the Method section, we examined the DTM questions within a Rasch (1960/1980) modeling framework. This approach will yield interval-level measurement of the construct, while simultaneously quantifying the model fit of items and persons along the same interval level dimension (called Outifit). Rasch scaling also identifies items that are answered in qualitatively different ways by subgroups of respondents (e.g., men vs. women) with equal DTM levels, so as to identify items that would bias the estimation of respondents’ level of DTM. Moreover, the Rasch model residuals can be factor analyzed to determine whether items define a unidimensional construct. By identifying the best fitting subset of unbiased items, Rasch scaling yields more realistic information regarding the distribution of DTM scores, as well as individual and group differences. We refer readers to Bond and Fox (2007) for a general introduction to Rasch scaling. Also, technical details and a thorough description of the parameter estimation technique can be found in Wright and Masters (1982).

Secondly, Blakemore et al.’s study relied exclusively on a sample of Western-based respondents. Thus, when administered to Chinese daters there is the distinct possibility that DTM scores are affected by cultural differences or by idiosyncratic features of the questions or their translations (see e.g., Lange et al., 2015; Lange, Thalbourne, Houran, & Lester, 2002). Rasch scaling can be used to assess potential biasing effects of factors like as age, sex, and education are studied in questionnaire design bias, and we interested readers can refer to our previous work for detailed discussions on the rationale and methodologies in question (Lange et al., 2002; Lange, Irwin, & Houran, 2000; Lange, Thalbourne, Houran, & Storm, 2000).

Participants

Data derived from a convenience sample of 3,389 paid subscribers (M_age = 30.4 yrs, SD = 8.2 yrs, range = 18 to 56 yrs.) to the Zhenai.com matchmaking service, consisting 1,780 men and 1,609 women. Respondents were not compensated for their participation. We note that this is not a representative sample of Chinese residents. As a matter of fact, the respondents were arguably not "traditionalists" in a social sense by the simple fact that they were seeking committed romantic relationships via the Internet. Therefore, our sample might best be characterized as “globalized” Chinese respondents (cf. Lange et al., 2015).

Measures and Procedure

We administered a 6-item modified, online version of Blakemore et al.’s (2005, pp. 329-330) Drive to Marry Scale that was translated into Chinese by bilingual members of Zhenai.com’s marketing team. The original scale was constructed using factor analysis, yielding five (5) items to be rated on a five-point scale (1 = disagree strongly; 5 = agree strongly) and yielding a Cronbach’s alpha of .86. The final scale score was a mean score of the five items.

To ensure consistency of the response format, our version omitted the only reverse-scored item in the original scale (“Getting married is not one of my top priorities”). Further, we expanded the scale with two (2) new items aimed at measuring more extreme trait levels. One item was a rating statement (“I am ready to do whatever is necessary to find a spouse”), and the other item was a fill in the blank question addressing expectation of marriage [“Once you find a suitable match, how long do you expect it will take for you to get married? ____ (months)”].

Next, a panel composed of marketers and matchmakers at Zhenai.com suggested altering the text of the original directions and response categories in order to be more aligned and relevant to Chinese culture: “Please tell if the below description is suitable for you” using four (4) categories: “Definitely Not Suitable, Somewhat Not Suitable, Somewhat Suitable, and Definitely Suitable.” We note that our previous scale developments and revisions have consistently utilized four categories with no neutral option.

Finally, in addition to the independent variables of respondent age, sex, education level, and marital status, we measured respondents’ social pressure (prescription) with a multiple choice inventory drawing on the Theory of Reasoned Action (i.e., “From your own point of view, which of the following sources think you should get married in the near future? (click all that apply): friends, family, colleagues at work/school, society at large, myself”). These particular sources of social influence were selected based on the practical insights from Zhenai.com matchmakers who have successfully paired Chinese singles for marriage.

Rasch Scaling

Rasch scaling assumes the existence of a latent trait that serves to quantify a respondent’s (t) trait level (T_i), together with the level (D) at which a question (j) address this trait. The parameter D_j is also called this item’s difficulty because it reflects items’ tendency to elicit lower vs. higher ratings, for higher vs lower and it is therefore also referred to as this item’s difficulty. Following the parameterization using in Bond and Fox (2007), the approach models the probability P_ijk – the probability that person i when faced with question j will select the (ordinal) answer k – based on the log odds:

where F_jk represents the item specific step values, i.e., the point at which for item i the ratings (k-1) and k are equally likely, and F₀ = 0 (see, e.g., Wright & Masters, 1982).

In the following we use Linacre’s (2013) Winsteps software to estimate the various parameters in Equation 1. Winsteps also quantifies model fit in terms of parameters’ “Outfit”, which is a standardized mean-square based index with a theoretical value of 1.0 (see, e.g., Wright and Masters, 1982). In general (Linacre, 2013), Outfit values below 1.4 are generally deemed acceptable. Note that in Equation 1 it is assumed that all variables contribute in an additive fashion, such that the variables exhibit the same pattern across other variables. Operationally this means that if we identify two (or more) sets of respondents with equal trait levels, then the D_j should be the same across groups. The finding of varying D_j across subgroups is also called Differential Item Functioning (DIF), and the difference between an item’s estimated D_j across groups quantifies DIF.

A check on DIF and dimensionality is also provided by a factor analysis of responses residuals, i.e., the difference between the actual ratings and those predicted by Equation 1. In contrast to standard factor analyses, model fit is supported by the absence of any clear factors. Winsteps can provide tests for DIF as well as analysis of the residuals.

Preliminaries

Consistent with expectations, the four items from Blakemore et al. (2005) and our two original items conformed to a unidimensional DTM scale because all items’ outfit values fell below the maximum criterion value of 1.4 (Linacre, 2013). The average person measure (i.e., the T_i in Equation 1) is 0.73 logits (SD = 1.70 logits), and the approximate overall reliability is 0.79. Table 2 shows the raw-score to Rasch translation in logits, the associated local SE, as well as more user-friendly scale scores with a mean of 100 and standard deviation of 10. In addition, the last column shows the local reliabilities, which range from 0.75 at the higher extreme to 0.94 near the middle of the dimension.

As is indicated by their (lower) logit values, Table 1 shows that items reflecting anticipation or urgency for marriage (ready to find a spouse, marriage is a major life goal) tend to precede items in the hierarchy that reflect feelings of pride and excitement regarding marriage (higher logit values reflect that items received lower ratings). The “easiest” item concerns respondents’ estimation of the number of months they expect it to take to get married once a suitable prospect has been found (at D = -1.05 logits). Please note that this value depends on the way the number of months was dichotomized here (12 months or fewer = 1, and 13 or more = 0).

As a further check we also performed a factor analysis of items’ residuals. The findings listed in Table 1 indicate that items’ loadings on the first residual factor range from -0.62 to 0.69, thus raising the possibility of item multidimensionality. However, the attenuation-corrected correlation among the three person measures (i.e., T in Equation 1) as computed over items with low, medium, or intermediate loadings all exceed 0.92. Therefore, we had strong justification to treat the items as unidimensional throughout the following.

Independent Variables

The following describes two type of analyses for each of the demographic and normative factors listed on the right side of Table 1. Each independent variable was dichotomized (where necessary) and coded as is described along the (bottom) row of the table labeled “Coded as High.’” This row indicates for instance that men were coded as “high” and “women” as low – and thus a positive value in the “Sex” column indicates that the item was harder for men than for women with equal DTM levels. We first describe a factor’s main effect on respondents’ overall DTM measure, thus revealing group differences with respect to the normative pressures faced by respondents. Secondly, we report items’ DIF associated with each factor.

Ancillary Analyses

The findings from the latter analyses suggest that the sources of the normative pressures themselves should be treated as if these were test items. The results of doing so are reported near the (right) bottom of Table 1 in the rows labeled “Social Referents: Item Locations” and “Social Referents: Outfit”). It can be see that “family is the “easiest” (i.e., most frequently occurring) social pressure (D = -0.99), followed by “self,” “friends,”, and “colleagues” (D = 0.20), while “society” is by far the hardest (i.e., least frequently occurring) social pressure (D = 1.40). At this point, we cannot recommend that these items should simply be added to the scale, the reasons being that (a) the items show considerable DIF, while (b) the Outfit of at least one source of pressure (“society”) is quite poor (Outfit = 2.12). Thus, further research is needed to address these issues.