Confirming Preferences or Collecting Data? Information Search Strategies and Romantic Partner Selection

Abstract

This article investigates two kinds of information search strategies in the context of selecting romantic partners. Confirmatory searching occurs when people ask for more information about a romantic partner in order to validate or confirm their assessment. Balanced searches are characterized by a search for risk information for partners rated as attractive and for attractiveness information about partners rated as risky in order to attain a more complete evaluation. A factorial survey computer program randomly constructed 5 types of partner descriptions and college-age respondents evaluated nine descriptions in terms of both health risk and romantic attractiveness outcomes. The results show little evidence of balanced search strategies: for all vignette types the respondents searched for attractiveness information. Regression analysis of the search outcomes showed no difference between males and females in the desire for attractiveness or risk information, the amount of additional information desired, or the proportion of descriptions for which more information was desired. However, an attractive physical appearance did increase the amount of additional information desired and the proportion of vignettes for which more information was desired. The results were generally inconsistent with a balanced search hypothesis; a better characterization of the respondents' strategy might be “confirmatory bias.”

In a pair of recent studies Fishbein and his colleagues investigated how young adults identify and evaluate social and behavioral cues relevant to the selection of a romantic partner (Fishbein et al., 2004; Hennessy et al., in press). The research was motivated in part by prior work indicating that persons believe they can assess romantic partner cues to maximize positive relationship outcomes while simultaneously identifying cues to personality characteristics and specific social behaviors that reduce their risk of HIV or STD infection (Gold et al., 1991; Keller, 1993; Maticka-Tyndale, 1991; Van derVelde et al., 1992, 1994; Williams et al., 1992). The use (but not efficacy) of these cues has been validated in experimental and survey studies (Agocha and Cooper, 1999; Clark et al., 1996; Dijstra et al., 2000) and the experimental evidence seems consistent with behavioral epidemiological data that show that persons engage in “safer sex” with sex partners they perceive as “risky” and do not engage in “safer sex” with sex partners that are perceived as “safe” (Fishbein and Jarvis, 2000; Gebhardt et al., 2003; Misovich et al., 1997; Montano et al., 2001; Peterman et al., 2000; Rhodes and Malotte, 1996).

In the first study (Fishbein et al., 2004) young adults evaluated 159 characteristics of a potential romantic partner in terms of romantic attractiveness and health risk. For example, they rated characteristics such as self-confident, happy with myself, dependent, is a good listener, has tattoos, or wants to spend quiet evenings together. The gamma between average risk and average attractiveness evaluations for the 159 aspects was –.61, suggesting that the more one is attracted to some aspect of a potential partner, the less likely is one to view that aspect as “risky.” A strength of the paper was the use of an extensive list of possible cues to risk and attractiveness. However, the design did not permit respondents to judge a potential partner's risk and attractiveness in the face of contradictory cues: the respondents were never asked to make a decision assessing attraction or risk when more than a single piece of information was available.

The second paper (Hennessy et al., in press) addressed this question via the factorial survey method by constructing descriptions of hypothetical romantic partners using both risky and attractive characteristics. There were four types of romantic partner descriptions: those consisting of highly attractive and low risk aspects, those of high risk and low attractiveness aspects, those first presenting highly attractive and low risk aspects but ending with high risk and low attractiveness aspects, and those first presenting high risk and low attractiveness aspects and ending with highly attractive and low risk aspects. All respondents evaluated 5 of each vignette type in terms of riskiness to one's sexual health, romantic attractiveness, likelihood of going on a date, of having unprotected sex, and of becoming infected with HIV or another STD if they did have unprotected sex.

Vignette type made a large difference in elevating or depressing the outcome evaluations. This was not particularly surprising in the extreme cases of high attractiveness/low risk and high risk/low attractiveness descriptions but the results implied that the differences between the “conflicting information” descriptions were attributable to a primacy effect (Hovland, 1958; Ohanian & Cunningham, 1987): descriptions that began with attractiveness information but ended with risk information were evaluated as more attractive than those that began with risk information and ended with attractive aspects. The opposite pattern was seen for the risk outcomes. Participant gender also predicted romantic partner evaluations. Males were less risk averse than females when confronted with risk dominated descriptions but were equivalent to females when evaluating attractiveness dominated ones. This is consistent with prior research on gender differences in reported sexual behavior (Hoyle et al., 2000).

The research reported here again uses a factorial survey approach, this time designed to investigate the information search methods used by the respondents after decisions about the potential romantic partner were made. Specifically, this paper investigates the use of confirmatory versus balanced information searches in the context of romantic and sexual partner selection. Confirmatory searching occurs when, after a judgment about a potential partner is made, respondents ask for more information in order to validate or confirm the initial judgment (Jonas et al., 2001). Thus, testing or challenging the initial hypothesis of risk or attraction based on additional data is avoided and searching for information consistent with the initial decision is the goal. In contrast, a balanced search strategy would be defined by searching for risk information about attractive partners and for attractiveness information about high risk partners in order to collect a more complete picture of the potential partner. In the context of personal advertisements or interactions on the internet, we might expect balanced information searches to be common given the assumption that persons looking for romantic partners tend to advertise their attractive characteristics and tend to avoid disclosing those that might lead to a negative initial reaction.

The present research also extends the earlier studies in three other ways. First, it includes an explicit physical “appearance” factor in the romantic partner descriptions. Other studies (e.g., Agocha and Cooper, 1999) have compared the effects of appearance (either via photographs or descriptions) to other features of hypothetical partners. By including an appearance factor in the descriptions used here the effects of physical appearance can be compared with those of personality and behavior. A second difference is that in this study, the mixed vignettes more evenly distribute the attractive and risky features in the description (see below for more details). This change may attenuate the order effect observed in the earlier studies because it reduces the difference between parts of the romantic partner description. Finally, we include here a non-informative vignette that was a constant for all respondents and had essentially no information relevant to either risk or attractiveness. A non-informative description is useful for determining what sort of information about potential partners is sought when there are no potentially biasing effects of previous information.

Research Questions

1. What are the effects of vignette type, gender and physical appearance on judgments of health risk, attractiveness and the likelihood of dating, having sex, and becoming infected?

A. Vignette type: Based on our earlier findings, vignettes with consistent attractiveness and risk information should produce the most disparate outcomes, whereas the inconsistent partner descriptions should produce intermediate values. Because our previous study showed that order of the information was important, we expect that inconsistent vignettes which begin with attractive characteristics will result in higher averages for attractive outcomes and in lower averages for risk outcomes compared to those which begin with risk characteristics. However, given the revised approach to presenting risk and attractiveness aspects in the same vignette the order effects found in the earlier paper may be reduced or eliminated. We are uncertain about the neutral descriptions. If they are truly information free, we might expect them to be difficult to assess in terms of risk or attractiveness and therefore receive values at about the median. However, because no risk aspects are used in their construction, neutral descriptions could be evaluated by the respondents as “relabeled” or “disguised” versions of attractive descriptions. In this case, they should be evaluated similarly to the attractive descriptions. Note that although the risk and attractiveness judgments are not the central concern of this study, they are important to analyze because they demonstrate whether or not the vignette types were effective in their role as treatment manipulations and because the revised format of the risk and attractiveness information may attenuate the differences between vignette types that we observed earlier. B. Gender: Because males are less risk averse than females, we expect that males will be more attracted to and be more likely to date and have sex with potential partners compared with females. At the same time, males will see potential partners as less risky and as less likely to infect them than will females. Based on our previous research however, we expect the differences between males and females to be greatest with respect to “risky” partners and smallest with respect to “attractive” partners. That is, we expect gender to interact with vignette type. C. Physical Appearance: We expect that both men and women will be more attracted to, and will be more likely to want to date and have sex with potential partners described as “attractive” than as “average” in appearance. More specifically, we expect that the more physically attractive the potential partner, the less the attention paid to “risky” characteristics. Thus we expect both a main effect of appearance as well as an interaction between appearance and vignette type.

2. What are the effects of vignette type, gender and physical appearance on information search strategies?

A. Vignette Type: Balanced search strategies should produce risk oriented searches for high attraction/low risk descriptions and attractiveness oriented searches for high risk/low attraction ones. A confirmatory search strategy should produce attractiveness oriented searches for high attraction/low risk descriptions and risk oriented ones for high risk/low attraction descriptions. If there is a primacy effect in evaluating the mixed vignettes, we would also expect primacy effects in information search. Again, it is uncertain what should occur in the neutral case. If balanced searches predominate, the search should be for a mix of attractiveness and risk aspects while if neutral descriptions are interpreted as disguised high attraction/low risk descriptions, a balanced search should produce risk searches. If confirmatory searches predominate, and neutral descriptions are interpreted as a disguised high attraction/low risk, there should be an attractiveness search. We don't know what to expect if neutral descriptions are interpreted as a distinct type under a model of a confirmatory search: because these descriptions are uninformative, it is difficult to determine what preference would be confirmed in this case.

B. Gender: Given that sensation seeking is positively correlated with male gender, we expect that males will search for more risky information than females. This hypothesis is based on past research on sensation seeking and partner selection (Henderson et al., 2005). We are uncertain however about the functional form of this effect – males may search for more risky information in all vignette types (producing a main effect) or just in inconsistent or high risk/low attraction one producing an interaction effect.

C. Physical Appearance. We have little experience relative to this question. Other research suggests (e.g., Dijstra et al., 2000) that appearance may override risk information when making initial decisions about behavioral risks and related outcomes, but how physical appearance affects information search is unknown.

Methods and Measures

Participants

Respondents were recruited to participate in the study on the campuses of two Philadelphia universities. The project was described to them and then necessary university IRB procedures in regard to informed consent were followed (i.e., respondents reviewed, signed, and returned a consent form that described the study and noted risks and benefits of participation). Each participant was compensated $10 for their time while responding to the survey which typically took 20–30 minutes to complete. We limited the analysis sample (N=393) to respondents between 18 and 28 years of age to be consistent with the earlier studies (Mean age = 20, Median age = 20, SD = 1.86). 46% were male. The respondents were predominantly Caucasian (48%) and African American (28%) with some Asian/Pacific Islanders (12%) and Hispanics/Latinos (4%). The remaining (8%) were of mixed or unknown ethnicity.

Survey Administration

The survey was computer administered using the software MediaLab (Jarvis, 1998.) The first section asked for age, gender, and sexual orientation so that the potential romantic partner was the appropriate gender for the respondent.[1] The second part presented the respondent with nine different vignettes using the factorial method. Respondents assigned each description a score for attractiveness, risk to their sexual health, the likelihood of going on a date, having sex, and becoming infected with an STD or HIV after unprotected sex. Then, as we describe below, twenty new characteristics reflecting risk, attractiveness, or their combinations were displayed and the respondent could choose up to five of them as additional informational items. The survey continued with demographic (e.g., race, ethnicity) and other individual difference questions relating to the respondent's behavior.

What is a Factorial Survey?

A factorial survey is a self-administered survey that presents hypothetical scenarios (i.e., vignettes) that are randomly constructed from mutually exclusive descriptive phrases. Respondents then make an evaluative judgment or a decision based on the data in the complete scenario. Factorial surveys are often used to model individual decision-making processes and consumer preferences. An example of a decision making survey is Hennessy, Manteuffel, DiIorio and Adame (1997) which modeled adolescent decisions to have sex on the basis of randomly constructed social contexts. Consumer preferences have been studied in respect to programs to control STD infection (Hennessy et al., 2002), HIV vaccine trials (Hennessy et al., 1996), and the desired features of STD/AIDs prevention programs (Hennessy et al., 2002). The construction and analysis of factorial surveys has been described in detail elsewhere (Hennessy et al., 1995; Hox et al., 1991; Rossi and Nock, 1982).

Variable Features of the Vignettes

There were five types of vignettes constructed. The constituent characteristics of the descriptions were selected based on results in Fishbein et al. (2004). Extreme examples at both ends of the implicit risk/attractiveness rating scale were selected to make sure that the romantic partner descriptions were extremely contradictory when both high risk/low attractiveness and high attractiveness/low risk descriptions were presented to the respondent. Four types of vignettes were randomly constructed from the attributes listed in Table 1:

Table 1.

Attributes and Levels Used to Construct the Romantic Partner Vignettes

ATTRACTIVENESS ATTRIBUTE

1. attractive

2. average

HIGH RISK ATTRIBUTE 1

1. is a smoker

2. uses drugs occasionally

HIGH RISK ATTRIBUTE 2

1. doesn't care about fitness

2. believes that more sex, the better

LOW ATTRACTIVENESS ATTRIBUTE 1

1. is often pessimistic

2. is agnostic in religious orientation

LOW ATTRACTIVENESS ATTRIBUTE 2

1. has tattoos

2. is dependent on others

3. is a high school graduate

HIGH ATTRACTIVENESS ATTRIBUTE 1

1. believes that sex should be saved for someone special

2. is flexible

HIGH ATTRACTIVENESS ATTRIBUTE 2

1. is self-confident

2. is trustworthy

LOW RISK ATTRIBUTE 1

1. is deliberate

2. likes to share every thought

LOW RISK ATTRIBUTE 2

1. is idealistic

2. takes initiative

3. is a team player

Note. Names for each potential romantic partner (ten for each gender) were also part of the random construction process, but the name of the potential partner played no role in the analysis or study. They were included to make the descriptions more realistic and less redundant. The names are not shown here.

(HiA-LoR): random selection from high attractiveness attributes 1 and 2 and from low risk attributes 1 and 2, presented in that order.

(HiR-LoA): random selection from high risk attributes 1 and 2 and low attractiveness attributes1 and 2, presented in that order

(MixedA): random selection from high attractiveness attribute 1, high risk attribute 2, low risk attribute 1, and low attractiveness attribute 2, presented in that order, and

(MixedR): random selection from high risk attribute 1, high attractiveness attribute 2, low attractiveness attribute 1, and low risk attribute 2, presented in that order.

There were, however symmetric restrictions placed on some of the selections to make the vignettes internally consistent: the high risk aspect “more sexual experience the better” could not appear with the high attractiveness aspect “believes that sex should be saved for someone really special” and the low attractiveness aspect “is often pessimistic” could not appear with the contradictory highly attractive aspect “is self-confident.”

As noted above, one additional vignette type was constructed, the Neutral (NEU) description. This consisted of only two aspects: the age aspect of “between 18 and 25” and the marital status aspect of “single.” This type was added to the mix of HiA-LoR, HiR-LoA, MixedR and MixedA to assess a “non-informative” condition because the NEU vignette contains virtually no useful discriminatory information about the potential partner because all potential romantic partners were described as “between 18 and 25” and “single”.

The physical appearance attribute also varied across types, but was not related to them. More specifically, after the five types of vignettes were constructed, one of the aspects of the physical appearance attribute (“attractive” or “average”) was randomly selected and this was always the last element of the constructed description.

In summary, HiA-LoR vignettes presented high attractiveness and low risk information while HiR-LoA types presented high risk and low attractiveness information. MixedA and MixedR combined aspects that represented both risky and unattractive and attractive and low risk aspects in different order while the NEU type supplied only minimal information about the romantic partner. The randomly selected physical appearance aspect was displayed last. Each respondent evaluated nine vignettes in all, two of the HiA-LoR, HiR-LoA, MixedA, and MixedR type and one NEU.

The Romantic Partner Descriptions

When the random selection, the constraints necessary for internal consistency, and gender specific pronouns were applied, a complete HiA-LoR description displayed for a heterosexual female, bisexual female, or homosexual male respondent might be:

Brian is single between the ages of 18 and 25. He is flexible and is trustworthy. He likes to share every thought and is a team player. Brian is average in appearance.

A HiR-LoA description for the same respondent might be:

Paul is single between the ages of 18 and 25. He is a smoker and doesn't care about fitness. He is often pessimistic and has tattoos. Paul is attractive in appearance.

A MixedR description for the same respondent might be:

Vance is single between the ages of 18 and 25. He is a smoker and is self confident. He is often pessimistic and is a team player. Vance is average in appearance.

A MixedA description for the same respondent might be:

Tony is single between the ages of 18 and 25. He believes that sex should be saved for someone special and doesn't care about fitness. He likes to share every thought and is a high school graduate. Tony is attractive in appearance

A NEU description for the same respondent might be:

Mark is single between the ages of 18 and 25. Mark is average in appearance.

The Outcome Measures

The respondents judged each potential partner vignette with respect to his/her potential sexual health risk, attractiveness, the likelihood of going on a date, the likelihood of having sex with the partner, and the likelihood that the respondent would become infected with an STD or AIDS from the partner after unprotected sex. The risk and attractiveness outcomes were scaled from 1 = extremely safe/extremely unattractive to 11 = extremely risky/extremely attractive and the likelihood items were scaled identically with “extremely unlikely” and “extremely likely” as the endpoints.

The last question about every vignette consisted of the stem “Given the information above, [i.e., the partner description] would you like to know if this potential partner:” where the options included 20 potential partner characteristics. These additional information aspects represented one of five types of aspects based on a cross tabulation of risk and attractiveness values attributed to them in an earlier study (Fishbein et al., 2004, Table IV). The 5 types and their informational aspects were:

Table 4.

Informational Aspects Desired Overall and by Vignette Type

		Type of Description
Informational Aspects (Code)	Total (N=3537)	HiR-LoA (N=786)	MixedR (N=786)	NEUTRAL (N=393)	MixedA (N=786)	HiA-LoR (N=786)
Is happy with his/her self (HiA-LoR)*	.487	.444	.424	.552	.527	.521
Is faithful (HiA-LoR) *	.426	.377	.416	.557	.403	.445
Is supportive of others (HiA-LoR)*	.324	.270	.300	.368	.333	.372
Wants a casual relationship (HiR-LoA)*	.292	.278	.301	.333	.262	.306
Likes to sexually experiment (R)*	.288	.286	.304	.335	.239	.300
Is a social drinker (A)*	.238	.181	.222	.259	.255	.284
Takes things as they come (A)*	.210	.169	.207	.231	.235	.221
Is a free spirit (A)*	.179	.153	.169	.201	.176	.208
Is athletic (A)*	.162	.099	.137	.198	.173	.223
Believes that life is short (R)	.159	.157	.178	.139	.150	.160
Wants to spend evenings together (HiA-LoR)*	.134	.085	.109	.137	.139	.201
Believes that cleanliness is over-rated (HiR-LoA)*	.120	.155	.108	.127	.125	.089
Is popular (UN)*	.111	.080	.101	.137	.115	.134
Feels unfulfilled (HiR-LoA)*	.110	.147	.128	.061	.113	.076
Is impulsive (R)	.108	.108	.111	.091	.114	.109
Is secretive/mysterious (R)*	.104	.096	.104	.165	.080	.106
Is a private person (UN)*	.102	.067	.096	.134	.113	.117
Has his/her head in the clouds (HiR-LoA)	.098	.105	.089	.086	.099	.106
Is eccentric (UN)	.095	.095	.094	.111	.090	.094
Has pierced body parts (UN)*	.065	.081	.072	.043	.067	.052

Spearman Correlations Among Proportions (N = 20)
	HiR-LoA	MixedR	NEU	MixedA
MixedR	0.910
NEU	0.758	0.879
MixedA	0.840	0.937	0.902
HiA-LoR	0.718	0.876	0.945	0.939

Note. Entries are the proportions of all romantic partner descriptions or type of description for which a particular informational aspect was selected. Rows are sorted by descending proportions of the total sample. The Code for each aspect reflects the informational content of each aspect based on earlier research. HiA-LoR is high attractiveness and low risk; A is high attractiveness only; UN is uninformative re: attractiveness or risk; R is high risk only; LoA-HiR is low attractiveness and high risk. Asterisks in aspect column indicate significant differences between vignette type for that particular aspect; all F tests are corrected for non-independence.

High Risk/Low Attractiveness: feels unfulfilled, has his/her head in the clouds, thinks that cleanliness is over-rated, wants a casual relationship.

High Risk/Neutral Attractiveness: likes to sexually experiment, is impulsive, is secretive/mysterious, believes that life is short and it should be lived to the fullest.

Neutral Risk/Neutral Attractiveness: is a private person, is eccentric, has pierced body parts, is popular.

High Attractiveness/Neutral Risk: is a social drinker, takes things as they come, is athletic, is a free spirit.

High Attractiveness/Low Risk: wants to spend evenings together, is faithful to friends and acquaintances, is supportive of others, and is happy with his /her self.

The 20 aspects were presented to the participant in three sets that were randomly ordered. The respondents could select from zero to five of the 20 aspects listed and could move forward and backward from screen to screen to reconsider their initial choices if necessary. The respondent's choices were used to create four additional variables in order to investigate the research questions related to information search. Two of the new variables were the number of additional informational aspects desired for each vignette and a dichotomous variable that assessed whether, for a given vignette, additional information was or was not desired. The other two new variables computed measured the type of information desired in the post-assessment search stage. Because we did not want to impose a bi-polar structure on the information search measure, one information search measure reflected a search for attractiveness information and the other a search for risk information. Specifically, high risk or high attractiveness aspects in the desired list were coded “1,” neutral aspects in the desired list as “0,” and low risk or low attractiveness aspects in the desired list as “−1.” The sum of the attractive or risk codes for the aspects selected (if any) when averaged by the number of items selected (e.g., 1–5) gives an average value for attractiveness and risk information desired. For the risk and attraction information search variables, positive values reflect a search for the characteristic (i.e., risky or attractive) and negative values the opposite.

Statistical Analysis

We use regression analysis to identify the higher order interactions between vignette type, physical appearance, and gender (although gender is not a design feature of the study, the random process that constructs the vignettes and then assigns them to respondents produces insignificant correlations between gender, appearance factor, and vignette type). For the continuous outcomes (the 5 partner assessments, the number of additional informational aspects wanted, and the two search type measures) we use random effects generalized least squares regression that adjusts the standard errors of the coefficients for the repeated measures design using the Huber-White estimator.[2] For the dichotomous outcome of wanting more information we use probit regression with the Huber-White correction applied and convert the Z-score metric coefficients (Agresti, 1990:102–104) into changes in the probability of wanting more information (Greene, 1993:639).

Results

Descriptive Statistics of the Assessments and Information Search Outcomes

Table 2 shows the means and standard deviations of the five judgments. Because the outcomes were scaled from 1–11, the midpoint value is 6. Most of the average values are slightly above the midpoint but one is far below (having sex). Looking at the information search summary statistics it can be seen that the average number of the five possible informational aspects requested was 3.82 (SD = 1.782) and respondents wanted more information for over 86% of all romantic partner descriptions. Over all potential romantic partner descriptions, the search was for attractive aspects (M = .405, SD= .394) while risk aspects were generally avoided (M = −.027, SD =.441).

Table 2.

Summary Statistics of Assessment and Information Search Outcomes

	N	Mean	Std.Dev.
Vignette Assessment Outcomes:
Attractiveness	3533	6.064	2.815
Go on a Date	3533	6.186	2.841
Have Sex	3516	4.226	2.973
Risk	3517	6.151	2.561
Get Infection	3471	6.366	2.444
Information Search Outcomes:
# of Aspects Wanted	3537	3.824	1.782
% Wanting More	3537	.868	.338
Attractiveness Search	3071	.405	.394
Risk Search	3071	−.027	.441

Note. The scale for all assessment outcome items is 1–11. Scale for # aspects wanted is 0–5. Percent wanting more is a dichotomy. Range of search outcomes are −1 to +1, computed only for respondents who desired additional information. The N here refers to the number of partner descriptions evaluated.

What are the effects of vignette type, appearance, and gender on judgments of risk and attractiveness and the other partner assessment outcomes?

Table 3 shows the ANOVA results for the five partner assessments. The table was produced the following way. All main and interaction effects were computed as dummy variables or products of dummy variables. This resulted in 19 parameter estimates representing the “saturated” model of all possible predictors (e.g., 4 main effects of vignette type, one main effect of gender, one main effect of appearance, four vignette type*appearance two way interactions, four vignette type*gender two way interactions, one gender*appearance two way interaction and four vignette type*gender*appearance three way interactions). Then all the interactions were excluded and the significance of the change in likelihood comparing the more complicated model (i.e., the saturated one) with the restricted model (i.e., the one with main effects only) was computed using a Wald test (Kennedy, 2003, pages 66–68). If this chi-squared value was not-significant, only the main effects were retained. If some main effects were then individually insignificant, they were deleted. However, if the initial test for the necessity of any interactions reported that they were necessary, we repeated the procedure evaluating the highest order (e.g., three-way) interaction first until the regression included all relevant sets of main or interaction dummy variables.

Table 3.

ANOVA Results for Partner Assessment Outcomes: Significant Main Effects and Interactions

Attractiveness Assessment (N = 3,533 vignettes, 393 respondents)
	χ2 (df)	p
Vignette	533.48(4)	p < .0001
Appearance	14.54 (1)	p < .0001
Gender	19.75 (1)	p < .0001
Appearance*Gender	7.44 (1)	p =.0064
Gender*Vignette	37.05 (4)	p < .0001
Appearance*Vignette	9.59 (1)	p = .048

Go on a Date Assessment (N = 3,533 vignettes, 393 respondents)
	χ2 (df)	p
Vignette	471.26(4)	p < .0001
Appearance	12.83 (1)	p = .0003
Gender	12.76(1)	p = .0004
Appearance*Gender	4.67 (1)	p = .0307
Gender*Vignette	37.16 (4)	p < .0001
Appearance*Vignette	9.76 (1)	p = .0446

Have Sex Assessment (N = 3,516 vignettes, 392 respondents)
	χ2 (df)	p
Vignette	533.49(4)	p < .0001
Appearance	48.40(1)	p < .0001
Gender	160.28 (1)	p < .0001
Gender*Vignette	18.02 (4)	p =.0009

Risk Assessment (N = 3,517 vignettes, 393 respondents)
	χ2 (df)	p
Vignette	749.25(4)	p < .0000
Gender	6.67(1)	p = .0098

Get Infected Assessment (N = 3,471 vignettes, 390 respondents)
	χ2 (df)	p
Vignette	510.24(4)	p < .0001

Note. All χ² tests are adjusted for the non-independence of observations. The saturated model included 3 main effects (vignette type, appearance factor, and gender) and all higher order interactions (19 df in all). χ² tests reflect the significance of the included factors in the final model.

Considering the assessment measures, all three main effects and all three two way interactions were significant for both the attractiveness and the going on a date outcomes (the Appearance*Vignette interaction just less than .05 in both cases). For the have sex assessment, all three main effects were significant as well as the gender*vignette interaction. For the risk assessment, there were only main effects of gender and vignette, and for the get infected assessment, only the vignette type had any discernable effect.

An analysis of the research questions related to equivalency of MixedR and MixedA and the HiA-LoR and NEU descriptions for the judgments requires significance tests on regression coefficients and their differences. To do this, we return to the estimated model in Table 4 and test specific hypotheses about differences between combinations of coefficients. Because all the terms in the equation are dichotomous, tests of differences of combinations are equivalent to tests of differences between predicted means. For the attractiveness and go on a date outcome, there are eight comparisons of means for each, 4 concerning the MixedR versus MixedA differences and four concerning the HiA-LoR versus NEU differences. Both outcomes show the same pattern: none of the comparisons are equivalent except for the MixedR versus MixedA comparison for females evaluating vignettes of average attractiveness (detailed results on these comparisons are available from the first author). For the risk assessment, the observed difference between the MixedR and MixedA is also discernable from chance (χ²= 97.64, p <.0001) as is the HiA-LoR and NEU difference (χ² = 223.68, p < .0001). For infection the same is true (χ² = 51.68, p < .0001 for MixedA versus MixedR difference; χ² = 172.91 p < .0001 for HiA-LoR versus NEU difference, all these χ² tests have a df equal to 1). The major exception to this general finding is the result for the have sex assessment. Here the two MixedR versus MixedA differences are non-significant as is the HiA-LoR versus NEU comparison for males. Female do differentiate between the HiA-LoR and NEU case when evaluating partner descriptions regardless of the attractiveness level of the description. Thus, there is still substantial evidence of a primacy effect of information when the MixedA and MixedR vignette types are compared. MixedR means were typically lower for attractiveness and going on a date and higher for outcomes like risk and getting infected when compared with MixedA means. Thus, even in vignettes in which positive and negative information is presented in a more balanced fashion, information presented first has greater impact on evaluative judgments than later information. Finally, the NEU vignettes appear to be considered more like a MixedA type than an HiA-LoR type for the assessment outcomes.

While the ANOVAs tell the statistical story, a better way to display the results is through plots. Figure 1 shows the predicted means, by vignette type, for the five assessment outcomes. The attractiveness and going on a data outcome require separate plots for male and female respondents and for vignettes that describe average or physically attractive potential romantic partners. For both of these outcomes, a physically attractive description increases the average levels of the outcome relative to romantic partners who were described as “average in attractiveness.” The gender*vignette type interaction shows that males evaluating HiR-LoA descriptions assess them as more attractive and “dateable” than females. With respect to having sex, attractive appearance increases the likelihood of having sex for both genders while males always give higher average assessments for this outcome than females, regardless of whether the romantic partner is described as average or attractive. For the risk outcome, males give lower risk assessments to type of partner descriptions relative to females, but both genders are sensitive to risk differences across types of vignettes (i.e., HiA-LoR have the lowest risk and HiR-LoA have the highest). For STD infection, only the vignette factor is necessary; there are no appearance or gender related differences.

Figure 1.

ANOVA Plots for Romantic Partner Assessments

What are the effects of vignette type, appearance, and gender on information search outcomes?

Now we shift the focus to the data on the twenty individual informational aspects - the choices of the information search task. Table 4 shows the proportion of vignettes for which each aspect was selected for the sample as a whole (column 2) and by vignette type (columns 3 through 7) as well as the Spearman correlations between the proportions. The order of the aspects in the table is determined by the proportion of the total cases where the respondent desired the indicated aspect. That is, for almost 50% of all vignettes, the respondents wanted to know more about whether the romantic partner described is “happy with him/her self” followed by whether the potential partners “is faithful” and “is supportive of others.” Note that the top four aspects are informative about both attractiveness and risk (see the discussion of the coding of this variable in the legend of Table 4). In contrast, the next most frequently selected group (from “likes to sexually experiment” to “believes that life is short”) are informative about attractiveness or risk, but not both. Finally, consider the four least selected characteristics. Three of the four are in the classification of “non-informative” about either attractiveness or risk (the fourth, “has his/her head in the clouds” – is informative, it is high risk and low attractiveness). Thus it appears that the respondents are searching initially for high attractiveness/low risk information, followed by those that are considered an indicator of at least one attribute. But the emphasis is clearly on attractiveness and not risk: features that reflect high attractiveness/low risk or attractiveness only dominate the top part of the preference list, while those that are uninformative or reflect high risk/low attractiveness or high risk only dominate the bottom of the preference list. Under a balanced search strategy high risk/low attractiveness features should be just as desirable to the respondent as high attractiveness/low risk features because they both provide information about attractiveness and risk. But there appears to be little evidence for a balanced strategy in this table.

For all but four characteristics, there were significant differences between vignette types in the desire to know about a particular feature. Another way of looking at vignette heterogeneity is through the correlation between the choices by vignette type. Although the Spearman rank order correlations (Table 4, bottom) between choice and description type are generally high, they are lowest between HiA-LoR and HiR-LoA types and highest between HiA-LoR and NEU, which suggests that NEU are being treated as HiA-LoR vignettes when information search is considered.

Table 5 shows the ANOVA results for the four information search outcomes. For both number of informational aspects desired and the probability of wanting more information, only the main effects of appearance and vignette type are necessary. For attractive searches, only vignette type makes a difference, but for risk searches, gender and vignette type are necessary.

Table 5.

ANOVA Results for Information Search Outcomes: Significant Main Effects and Interactions

Number of Additional Aspects Desired (N = 3,537 vignettes, 393 respondents)
	χ2 (df)	p
Vignette	102.85(4)	p < .0001
Appearance	9.78 (1)	p = .0018

% Wanting Additional Information (N = 3,537 vignettes, 393 respondents)
	χ2 (df)	p
Vignette	81.85(4)	p < .0001
Appearance	5.56(1)	p = .0184

Attractiveness Searches (N = 3,071 vignettes, 390 respondents)
	χ2 (df)	p
Vignette	59.09(4)	p < .0001

Risk Searches (N = 3,071 vignettes, 390 respondents)
	χ2 (df)	p
Vignette	14.95(4)	p < .0001
Gender	8.19(1)	p = .0042

Note All χ² tests are adjusted for the non-independence of observations. The saturated model included 3 main effects (vignette type, appearance factor, and gender) and all higher order interactions (19 df in all). χ² tests reflect the significance of the included factors in the final model.

The regression analysis applied to the information search outcomes also shows a pattern of significant differences between the MixedR and MixedA descriptions for the information wanted, percent wanting more information, and both kinds of information searches (for information wanted, χ² = 4.38, p = .0368; for % information wanted, χ² = 4.89, p = .0271; for attractive searchers, χ² = 9.59 p = .0020; and for risk searches, χ² = 414.02, p = .0002, all these tests have one df). However, only for the information wanted outcome are HiA-LoR vignettes differentiated from NEU ones (χ² = 5.73, p = .0166). Thus for information search outcomes, the ordering of the information in the mixed descriptions remained important, but in three of the four cases, the NEU vignettes were treated as HiA-LoR types.

Again, plots of the significant effects are more informative than tests of ratios of sums of squares. These plots are found in Figure 2. Partner attractiveness increases the desire for more information for all types of descriptions, searches for attractiveness information are the norm for all types of descriptions, and males search for more risky features compared to females while both genders have higher risky search values for HiR-LoA and MixedR vignette types than for the other three. All these results are inconsistent with balanced information search strategies.

Figure 2.

ANOVA Plots for Information Search Outcomes

Discussion

Vignette type had strong effects on all assessments made about the hypothetical romantic partners. As expected, HiA-LoR vignettes showed highest average values for attractiveness, going on a date, and having sex, and HiR-LoA vignettes the highest average values for negative outcomes like risk and infection. The mixed types were intermediate with primacy effects present, except with respect to having sex. In general, the NEU type was interpreted as a mixed attractiveness (MixedA) vignette, not as a disguised HiA-LoR type.

In general, the results highlight the contextual nature of romantic partner preferences elicited. It's clear that the relationship between vignette type, gender, appearance and potential partner assessments is not a simple one. For positive outcomes like attractiveness and dating, both the appearance factor and respondent gender were important in addition to the risk/attractiveness aspects that constituted the description. Consistent with previous research, males are more attracted to, and are more likely to want to date and to have sex with potential partners than are females. However, the impact of gender on attraction and dating was moderated by the type of vignette. Men and women differed with respect to HiR-LoA vignettes but not with respect to HiA-LoR vignettes. Perhaps the most striking finding was that physical appearance significantly influenced positive but not negative outcomes. Both men and women were more likely to be attracted to, want to date, and want to have sex with, potential partners who were attractive in appearance than those who were “average in appearance.” Gender did interact with appearance, but only with respect to the attractiveness assessment: men's judgments of attraction are significantly more influenced by physical appearance than are women's. When having sex is considered, men are more likely to have sex with all partner types and, for both genders, an attractive description elevated the probability of having sex. Very surprisingly, and in contrast to expectations based on past research, physical appearance did not influence judgments of risk or the likelihood that one could get infected if one had unprotected sex with the potential partner. To put this somewhat differently, although men are consistently less risk averse than are women, neither men nor women reduced their perceptions of risk when confronted with a physically attractive partner. In summary, these respondents seem to have a relatively sophisticated process of romantic partner selection. For attractive but not for risky judgments, appearance plays a role. Moreover, the type of description is always relevant to the respondents irrespective of whether the partner is described as attractive or average in appearance. Research hypotheses in the future need to be much more complicated and sensitive to contextual information.

When we look at information search, the results are less complex. Information search was related to vignette type: in 79% of the HiR-LoA vignettes the respondents desired more information compared with 93% of the HiA-LoR types. Another way to show the differences between vignette types is to correlate the risk and attractiveness ratings of each with the number of additional information aspects desired. The higher the risk ratings, the lower the number of additional aspects desired (r = −.14, p < .05) and the higher the attraction rating, the greater the number of additional aspects desired (r = .24, p < .05). The signs of these correlations are inconsistent with a balanced search hypothesis. In fact, the average additional information desired was attractiveness related rather than risk related for all types of vignettes although in HiA-LoR vignettes the tendency for attractiveness related search was the highest and in the HiR-LoA type the search for attractiveness aspects was the lowest. All of these results are contrary to what would be expected under the balanced search hypothesis. Thus, there seems little evidence for a balanced strategy.

Gender plays no role in predicting information search outcomes. Thus, the hypothesis of gender differences in search patterns was not supported. However, appearance was important for information searching, but not for the type of information that was desired. In fact, the results show a general desire for more information and this tendency (both in terms of the average number of information aspects desired and the proportion of all vignettes for which more information is desired) was greater for vignettes describing an “attractive” potential partner as compared with one merely “average” in appearance. Finally, MixedR and MixedA descriptions were not treated similarly for any of the search outcomes, while NEU and HiA-LoR were in three of four cases.

In general, compared with romantic partner assessments, which is a complicated task, the information search strategy used by the respondents is relatively simple. Perhaps a better label for the information search process revealed here is one of “confirmation bias” (Nickerson, 1998) because searches for risk information are rare in all but HiR-LoA and MixedR cases. But even this result can be interpreted as a test of the “this is a potential high risk partner” hypothesis and is just another facet of confirming a pre-conceived assessment. Under a balanced search rule, information search should be for attractive features in the HiR-LoA case and clearly it is not. Confirmation bias is also demonstrated by the results for the HiA-LoR and the NEU vignettes. In terms of searching for both attractive and risk information, respondents treated these types similarly and searched more for attractive aspects within these types than any other. The high correlation (.945) between the specific type of informational aspect desired for HiA-LoR and NEU vignettes is another way of demonstrating their essential similarity (see Table 2, bottom). The regression analyses of the search outcomes confirm that the average search type value is the same for NEU and HiA-LoR outcomes and the number of additional informational aspects desired is actually highest in NEU, followed by HiA-LoR. Final evidence for the operation of confirmation bias comes from the analysis of the informational aspects that were selected from the list of twenty: they emphasized high attractiveness and not high risk. Note that the specific informational aspects that were desired tend to emphasize personality traits that probably would not be evident given a casual encounter (e.g., happy with his/herself, faithful, supportive of other, wants a casual relationship, see Table 3).

Why is confirmatory searching the general strategy? Romantic partner selection is one decision context where the perils of false positives (e.g., selecting someone apparently supportive of others, but is also a promiscuous drug user) generally outweigh the lost opportunities represented by false negatives (e.g., rejecting a smoker who is also happy with his/herself, faithful, and wants to spent quiet evenings together). The first case is a worst mistake than the second, especially if it takes time to find out the reality of the situation. As Nickerson, (1998, page 204) summarizes this situation abstractly: “When, for example, the undesirable consequences of judging a true hypothesis to be false are greater than those of judging a false hypothesis to be true, a bias toward confirmation is dictated by some normative models of reasoning and by common sense.”

Regardless of the explanation for confirmation bias, partner search strategies based on confirmatory rules are likely to be the most risky in terms of health consequences because they actively avoid a systematic hunt for risk information or for mixed information that may nonetheless be risk informative. Insofar as confirmatory searches are usually biased against discovering negative or risky information, they could lead to unsafe sex with an apparently “safe” partner because relevant risky information was not revealed. However, there is evidence that training to identify and avoid biasing or incorrect decision-making heuristics can be successful (e.g., Moutier & Houdé, 2003; Erb et al., 2002; Nickerson, 1998, page 211), so it may be feasible to design interventions targeted at high school and college-aged youth where the use of confirmatory strategies for information searches could be identified and participants could be trained to use a information search heuristic that emphasizes a balanced rather than confirmatory orientation.

This study has three weaknesses. First, the use of a convenience sample limits the potential generalizabilty of the results. In contrast, what would be ideal is a web-based version of the questionnaire and vignette program to which representative samples of respondents, both adolescents and adults, could be connected. Second, like all vignette type approaches, the hypothetical nature of the assessment and information search process probably does not correspond exactly to the assessment and information search processes used “in the field” by adolescents and adults. The next step in our series of studies is, in fact, a small study of risk and attractiveness assessments made by respondents of their actual romantic partners. Our interest is in attempting to replicate the “vignette types” through classifications of actual romantic partners and their perceived attributes to determine if risk and attractiveness assessments of actual partners are similar to those of hypothetical vignettes. Finally, a third weakness of this study is the number of still unresolved questions it raises. For example, the lack of gender differences in information search is an interesting finding and one that is inconsistent with what we know about gender differences in partner preferences and risk reaction. But should information search be expected to be different by gender, and if so, what is the rationale for this expectation? This is only one of many research questions that remain unanswered.