Extracting multistage screening rules from online dating activity data @article{BruchExtractingMS, title={Extracting multistage screening rules from online dating activity data}, author={E. Bruch and F. Feinberg and Kee Yeun Lee}, journal={Proceedings of the National Academy of Sciences}, year={}, volume={}, pages={ - } } This paper presents a statistical framework for harnessing online activity data to better understand how people make decisions. Building on insights from cognitive science and decision theory, we develop a discrete choice model that allows for exploratory behavior and multiple stages of decision making, with different rules enacted at each blogger.com by: 17 30/08/ · Extracting multistage screening rules from online dating activity data. Elizabeth Bruch Department of Sociology, University of Michigan, Ann Arbor, MI ; Center for the Study of Complex Systems, University of Michigan, Ann Arbor, MI ;
Extracting multistage screening rules from online dating activity data. - Abstract - Europe PMC
Either your web browser doesn't support Javascript or it is currently turned off. In the latter case, please turn on Javascript support in your web browser and reload this page.
Proceedings of the National Academy of Sciences of the United States of America30 Aug38 : DOI: Free to read. b Center for the Study of Complex Systems, University of Michigan, Ann Arbor, MI. e Department of Management and Marketing, Hong Kong Polytechnic University, Kowloon, Hong Kong.
Author contributions: E. designed research; E. performed research; E. and F. analyzed data; and E. wrote the paper. Online activity data—for example, from dating, housing search, or social networking websites—make it possible to study human behavior with unparalleled richness and granularity. However, researchers typically rely on statistical models that emphasize associations among variables rather than behavior of human actors.
Harnessing the full informatory power of activity data requires models that capture decision-making processes and other features of human behavior. Our model aims to describe mate choice as it unfolds online. It allows for exploratory behavior and multiple decision stages, with the possibility of distinct evaluation rules at each stage.
This framework is flexible and extendable, extracting multistage screening rules from online dating activity data, and it can be applied in other substantive domains where decision makers identify viable options from a larger set of possibilities.
This paper presents a statistical framework for harnessing online activity data to better understand how people make decisions. Building on insights from cognitive science and decision theory, we develop a discrete choice model that allows for exploratory behavior and multiple stages of decision making, with different rules enacted at each stage.
Critically, the approach can identify if and when people invoke noncompensatory screeners that eliminate large swaths of alternatives from detailed consideration. The model is estimated using deidentified activity data on 1.
A nonparametric account of heterogeneity reveals that, even after controlling for a host of observable attributes, mate evaluation differs across decision stages as well as across identified groupings of men and women. Vast amounts of activity data streaming from the web, smartphones, and other connected devices make it possible to study human behavior with an unparalleled richness of detail.
Taking full advantage of the scope and granularity of such data requires a suite of quantitative methods that capture decision-making processes and other features of human activity i. Discrete choice models, by contrast, can provide an explicit statistical representation of choice processes. However, these models, as applied, often retain their roots in rational choice theory, presuming a fully informed, computationally efficient, utility-maximizing individual 1. Over the past several decades, psychologists and decision extracting multistage screening rules from online dating activity data have shown that decision makers have limited time for learning about choice alternatives, limited working memory, and limited computational capabilities.
As a result, a great deal of behavior is habitual, automatic, or governed by simple rules or heuristics. For example, when faced with more than a small handful of options, people engage in a multistage choice process, in which the first stage involves enacting one or more screeners to arrive at a manageable subset amenable to detailed processing and comparison 2 — 4.
These screeners eliminate large swaths of options based on a relatively narrow set of criteria. Researchers in the fields of quantitative marketing and transportation research have built on these insights to develop sophisticated models of individual-level behavior for which a choice history is available, such as for frequently purchased supermarket goods. However, these models are not directly applicable to major problems of sociological interest, like choices about where to live, what colleges to apply to, and whom to date or marry.
To that end, here, we present a statistical framework—rooted in decision theory and heterogeneous discrete choice modeling—that harnesses the power of big data to describe online mate selection processes.
Our approach allows for multiple decision stages, with potentially different rules at each. We apply our modeling framework to mate-seeking behavior as observed on an online dating site.
In doing so, we empirically establish whether substantial groups of both men and women impose acceptability cutoffs based on age, height, body mass, and a variety of other characteristics prominent on dating sites that describe potential mates.
The pool of potential partners includes all relevant users active on the site. Informative features of mate choice behavior are revealed at each stage, and choices made at the browsing stage restrict which alternatives are subsequently available. One may, for example, browse a narrow band of ages and then be relatively indifferent to age thereafter when writing. Empirical studies suggest that the extracting multistage screening rules from online dating activity data process commences using cognitively undemanding, cutoff-based criteria operating on a small number of attributes e.
Our proposed framework can accommodate an arbitrary number of sequentially enacted winnowing stages. Here, extracting multistage screening rules from online dating activity data, we focus on two intrinsic to the medium: browsing and writing. At each stage, choice is governed by one or more possible decision rules, extracting multistage screening rules from online dating activity data, which are uncovered by the model.
Alternately, extracting multistage screening rules from online dating activity data, they may impose noncompensatory screening rules, in which they browse only those profiles meeting some threshold of acceptability on one or more attributes.
Decision theorists distinguish screeners that are conjunctive deal breakers from those that are disjunctive deal makers ; the former indicates a set of qualities where all must be possessed, and the latter indicates a set of qualities where any one suffices.
Even sophisticated modeling approaches in social research 7extracting multistage screening rules from online dating activity dataalthough offering great flexibility to fit data well, extracting multistage screening rules from online dating activity data, typically encode two procedures at odds with how actual humans seem to process large amounts of information.
However, noncompensatory decision rules that allow for i abrupt changes in the relative desirability of potential partners as an attribute passes outside an acceptability threshold and ii an attribute to have a disproportionate effect on choice outcomes over some region of values lack anything approaching a turnkey solution. Such splines consist of linear functions joined at specific points called knots. If knot positions are known in advance—for example, a downturn in utility for men under a given height—estimating the slopes of each of the component linear functions is straightforward and quick; however, here, we seek to identify both the slopes and the knots themselves, which are highly nontrivial The key impediment to efficient estimation is that the space of all possible knots is typically very large for our final model, on the order of 10 62 in factand therefore, brute force exhaustive search is out of the question.
Thus, one needs a powerfully efficient way to explore potential knot configurations Materials and Methods. With V on a logit scale, a difference of three represents a difference in odds and thereby, probability on the order of being 20 times less likely that the potential match will be browsed or written to, which may be large enough that no other attribute combination can overcome it: a deal breaker.
Illustration of how choice model captures alternative decision rules. A depicts a linear compensatory rule; B depicts a nonlinear but compensatory one. C is a conjunctive rule where being outside of the range δ 1 i k and δ 2 i k acts as a deal breaker, and D is a disjunctive rule where being greater than δ 2 i k acts as a deal maker. In summary, the model accommodates three key constructs: i nonlinear, even noncompensatory, evaluative processes; ii heterogeneity across individuals; and iii multistage choice behavior.
For our specific application to online dating, it allows for distinct but statistically intertwined accounts of both the browsing and writing stages and explicit quantification of the relative importance placed on observable attributes included in online profiles. The model also accommodates exploratory and stochastic behavior, thus guarding against a deal breaker on, say, age being tautologically inferred as the oldest or youngest value observed for each individual. Our data consist of over 1.
For categorical attributes, dummies capture potential interactions. Differences likely matter more at low vs. Both BMI and age are, therefore, accommodated as differences on a log scale [e. Table 1 reports the fits of two-stage models with and without heterogeneous decision rules latent classes as well as models that allow for conventional representation of continuous covariates i.
Based on standard fit metrics [Bayesian Information Criterion BIC and L 2 ], the proposed model with five latent classes for both men and women fits the data better than all nested models e. To safeguard against overfitting, we also assess goodness of fit using a holdout sample consisting of men and women who joined the site immediately after the estimation period. These out of sample estimates reaffirm that a model allowing for nonsmooth response and heterogeneity outperforms other more traditional specifications.
In addition to superior fit, our model captures features of decision processes that are distorted by traditional approaches. Additional details are in SI AppendixSection S4. Although our models produce many results, we focus here on key features of mate choice behavior that would be, as a whole, inaccessible with alternative modeling approaches: i different rules at different decision stages, ii sharp cutoffs in what attribute values are desired or acceptable, iii invocation of deal breakers, and iv heterogeneity in behavior.
All results reported in the main text are significant at the 0. Distinct subsets of attributes are implicated at the browsing and writing stages. For example, when men select among women, age plays a greater role in the browsing stage.
Consider Fig. Among women, age matters in both browsing and writing, but its effects can vary across stages. For example, as we see in Fig. BMI also figures differently into browsing and writing decisions. Thus, it seems that women can never be too thin to write to; conditional on browsing. Extracting multistage screening rules from online dating activity data probability of browsing and writing someone of a given value of age relative to the probability of browsing or writing someone of equal age.
The y axis shows the associated probability ratio for both browsing and writing. The probability of browsing and writing someone of a given value of body mass relative to the probability of browsing or writing someone of equal body mass. The y axis shows the associated probability ratio.
By identifying sharp cutoffs in acceptability criteria, the model can identify norms or rules that would be difficult to extract using traditional methods. The results for height, as shown in Fig. Overall, women seem to prefer men who are 3—4 in taller across the board, with substantial drop offs for men below this cutoff.
This finding is consistent with prior research showing that women prefer a partner who is not taller than she is in heels With regard to age Fig. Any such crisp criteria would be smoothed over in a model that captured nonlinearities via polynomial specifications. The probability of browsing and writing someone of a given value of height relative to the probability of browsing or writing someone of equal height.
The x axis is height difference in inches between the user and potential match. Age differences are the biggest deal breaker. Even within the bulk of observations i. The model can also locate deal breakers in categorical covariates, although this is not unique to its framework. In online dating, one that stands out is not demographic but an act of omission: failing to provide a photo.
Both men and women are roughly 20 times less likely to browse someone without a photo, even after controlling for all other attributes in the model age, education, children, etc. Nearly as strong is smoking behavior: among those who do, nonsmokers are nearly 10 times less likely to be browsed and, therefore, smoking is evidently a decisive screen, extracting multistage screening rules from online dating activity data.
In short, we find clear evidence of deal-breaking behavior, although the strength of effects varies across the revealed classes.
Note that, although none of these may be truly inviolable, they are practically insurmountable within the observed range of available covariates. By allowing for unobserved heterogeneity, we can both assess what behaviors hold across the board and identify subclasses extracting multistage screening rules from online dating activity data users pursuing unique mate extracting multistage screening rules from online dating activity data strategies.
This paper presents a statistical framework for harnessing online activity data to better understand how people make decisions. Building on insights from cognitive science and decision theory, we develop a discrete choice model that allows for exploratory behavior and multiple stages of decision making, with different rules enacted at each blogger.com by: 17 Request PDF | Extracting multistage screening rules from online dating activity data | This paper presents a statistical framework for harnessing online activity data to better understand how Extracting multistage screening rules from online dating activity data @article{BruchExtractingMS, title={Extracting multistage screening rules from online dating activity data}, author={E. Bruch and F. Feinberg and Kee Yeun Lee}, journal={Proceedings of the National Academy of Sciences}, year={}, volume={}, pages={ - } }
Keine Kommentare:
Kommentar veröffentlichen