The Link Between Age and Partner Preferences in a Large, International Sample of Single Women

Women’s capacity to reproduce varies over the life span, and developmental goals such as family formation are age-graded and shaped by social norms about the appropriate age for completing specific developmental tasks. Thus, a woman’s age may be linked to her ideas about what an ideal partner should be like. With the goals of replicating and extending prior research, in this study we examined the role of age in women’s partner preferences across the globe. We investigated associations of age with ideal long-term partner preferences in a cross-cultural sample of 17,254 single (i.e., unpartnered) heterosexual women, ages 18 to 67, from 147 countries. Data were collected via an online questionnaire, the Ideal Partner Survey. Confirming our preregistered hypotheses, we found no or only negligible age effects on preferences for kindness-supportiveness, attractiveness, financial security-successfulness, or education-intelligence. Age was, however, positively associated with preferences for confidence-assertiveness. Consistent with family formation goals, age was associated with an ideal partner’s parenting intentions (high until approximately age 30, then decreasing afterward). Age range deemed acceptable (and in particular, the discrepancy between one’s own age and the minimum ideal age of a partner) increased with age. This latter pattern also replicated in exploratory analyses based on subsamples of lesbian and bisexual women. In summary, age has a limited impact on partner preferences. Of the attributes investigated, only preference for confidence-assertiveness was linked with age. However, age range deemed acceptable and an ideal partner’s parenting intention, a dimension mostly neglected in earlier research, substantially vary with age. Supplementary Information The online version contains supplementary material available at 10.1007/s12110-023-09460-4.


Deviations from Preregistration
Our hypotheses and methods were preregistered on the Open Science Framework: https://osf.io/qe3dr/.We deviated in our preregistration in several instances.These are mentioned as footnotes throughout the main manuscript and described in more detail in Table S1.

Footnote
Components in preregistration

Details of and rationale for the deviation
How might the deviation affect the results?
2 Hypotheses on parenting intention The preregistration did not include hypotheses for linear effects of age on importance of shared preference for number of children or preferred level of partner's intention to become a parent for models including a quadratic effect of age.For the sake of completeness and accuracy we include specific hypotheses about linear effects that are in line with linear hypotheses H6a) and H6b).
Hypotheses are in line with the other hypotheses that were specified in the preregistration.

Languages of the Ideal Partner Survey
The preregistration reports only nine language versions.Italian is erroneously missing from this list.1,265 of 17,254 women in our main sample filled out the survey in Italian (221 of 3,553 in our exploratory analyses).
Excluding these women from our analyses would have greatly reduced the sample size and seemed disproportionate given that we did not expect associations in Italian speaking women to differ from the rest of the sample.None of the robustness checks led to diverging results, therefore the decision to include women with missing answers to the seriousness question likely did not affect results.Both approaches (potentially including unserious women in the main analyses and potentially excluding serious women in the robustness analyses) may have lowered the ability to detect small effects (the first approach by introducing random noise, the second approach by reducing sample size).

Details of and rationale for the deviation
How might the deviation affect the results?

Model convergence
The preregistration states that we would include a random intercept and slope for country.As all main models did not converge when including a random slope for country, all analyses (including robustness and exploratory analyses) were performed only with a random intercept for country/ language.This procedure was described in the preregistration but to keep the main article short and precise we did not include the detailed model convergence procedure in the main manuscript.
As this is not a real deviation from the preregistration but rather a shortened description of the procedure with model convergence it did not affect the results.
Note.This deviation form was inspired by the Preregistration Planning and Deviation Documentation (PPDD) template: https://osf.io/ywrqe/( Van't Veer et al., 2019).Importance rating for shared preference for number of children by age applying the two-lines approach by Simonsohn (2018) Note.Graph and underlying analysis are based on R code provided at http://webstimate.org/twolines/.Visuals are slightly adapted to enhance readability.

Fig. S3
Preferred level of partner's intention to become a parent by age applying the two-lines approach by Simonsohn (2018) Note.Graph and underlying analysis are based on R code provided at http://webstimate.org/twolines/.Visuals are slightly adapted to enhance readability.

Exclusion Steps
For our exploratory analyses for lesbian and bisexual women we excluded the following participants: (1) all participants who did not identify as women; (2) all participants who did not identify as lesbian/gay/homosexual or bisexual/pansexual; (3) all participants who indicated being in a relationship or where it was not certain if participants were currently single; (4) all participants who did not answer the survey seriously or chose not to answer the seriousness question 1 .This led to sample sizes of 467 women identifying as lesbian and 3,085 women identifying as bisexual.Exclusion steps for our exploratory analyses are illustrated in Fig. S4.
1 Due to a technical error the answer to this question was missing in 369 cases in the sample of bisexual women and in 41 cases in the sample of lesbian women.The preregistration did not mention what to do in case of missing answers.Since the technical error was random we decided to keep the individuals for our analyses.In addition, we performed extensive robustness checks.See footnote 6 in Table S1 for further information.

Results Exploratory Analyses Based on Lesbian Women
Note.All analyses included a linear effect of age as a predictor and a random intercept for country.Sample size and number of countries differ for reasons explained in detail in the section "Sample Characteristics" of the main manuscript.For all hypotheses the raw and the standardized beta coefficient for the linear effect of age are displayed.The effect of age never reached substantiality (p < .005and |β| > 0.10 with CIs excluding ±0.10).b = raw beta coefficient, β = standardized beta coefficient, CI = 99.5% confidence interval.Note.Models for hypotheses H6a and H6b included age as a linear predictor and a random intercept for country.Models for hypotheses H7a and H7b included age as a linear and as a quadratic predictor and a random intercept for country.Sample size and number of countries differ for reasons explained in detail in the section "Sample Characteristics" of the main manuscript.For all hypotheses except for H7a and H7b the raw and the standardized beta coefficient for the linear effect of age are displayed.For H7a and H7b the raw and the standardized beta coefficient for the quadratic effect of age are displayed.In the models including only the linear predictor, the effect of age never reached substantiality (p < .005and |β| > 0.10 with CIs excluding ±0.10).b = raw beta coefficient, β = standardized beta coefficient, CI = 99.5% confidence interval.

Fig. S6 Effect size estimates of linear and quadratic age effects on preference for parenting intention for main analyses and three robustness analyses in the sample of lesbian women
Note.Robustness analyses 1: random intercept for language instead of countries; robustness analyses 2: excluding women with missing answers to the seriousness question (nexcluded = 41); robustness analyses 3: random intercept for language instead of countries and excluding women with missing answers to the seriousness question (nexcluded = 41).Graphs display standardized linear and quadratic age effects, error bars represent 99.5% confidence intervals.
Confidence intervals for effects estimates based on linear models are narrower compared to confidence intervals for effect estimates based on quadratic models because including a quadratic term decreases accuracy of estimates.In the models including only the linear predictor, the effect of age never reached substantiality (p < .005and |β| > 0.10 with CIs excluding ±0.10).

Fig. S7
Preference for parenting intention by a linear and quadratic effect of age controlled for a random intercept for country in the sample of lesbian women Note.Parenting intention includes importance rating for a partner sharing the preference for number of children (H7a) and preference for the level of partner's intention to become a parent (H7b).Blue areas represent 99.5% confidence intervals..01 Note.All analyses included a linear effect of age as a predictor and a random intercept for country.Sample size and number of countries differ for reasons explained in detail in the section "Sample Characteristics" of the main manuscript.For all hypotheses the raw and the standardized beta coefficient for the linear effect of age are displayed.Bold effect size estimates indicate substantial effects (p < .005and |β| > 0.10 with CIs excluding ±0.10).b = raw beta coefficient, β = standardized beta coefficient, CI = 99.5% confidence interval.

Fig. S8
Effect size estimates of linear effects on age-range deemed acceptable and on youngest and oldest age deemed acceptable for main analyses and three robustness analyses in the sample of lesbian women Note.Robustness analyses 1: random intercept for language instead of countries; robustness analyses 2: excluding women with missing answers to the seriousness question (nexcluded = 41); robustness analyses 3: random intercept for language instead of countries and excluding women with missing answers to the seriousness question (nexcluded = 41).Graphs display standardized linear age effects, error bars represent 99.5% confidence intervals.Dotted lines indicate the inference criteria for substantial effects (p < .005and |β| > 0.10 with CIs excluding ±0.10).

Fig. S9 Age-range deemed acceptable and youngest and oldest age deemed acceptable by own age in the sample of lesbian women
Note.Graphs display means and 99.5% confidence intervals.Note.Models for hypotheses H6a and H6b included age as a linear predictor and a random intercept for country.Models for hypotheses H7a and H7b included age as a linear and as a quadratic predictor and a random intercept for country.Sample size and number of countries differ for reasons explained in detail in the section "Sample Characteristics" of the main manuscript.For all hypotheses except for H7a and H7b the raw and the standardized beta coefficient for the linear effect of age are displayed.For H7a and H7b the raw and the standardized beta coefficient for the quadratic effect of age are displayed.In the models including only the linear predictor, the effect of age never reached substantiality (p < .005and |β| > 0.10 with CIs excluding ±0.10).b = raw beta coefficient, β = standardized beta coefficient, CI = 99.5% confidence interval.

Fig. S11
Effect size estimates of linear and quadratic age effects on preference for parenting intention for main analyses and three robustness analyses in the sample of bisexual women Note.Robustness analyses 1: random intercept for language instead of countries; robustness analyses 2: excluding women with missing answers to the seriousness question (nexcluded = 369); robustness analyses 3: random intercept for language instead of countries and excluding women with missing answers to the seriousness question (nexcluded = 369).Graphs display standardized linear and quadratic age effects, error bars represent 99.5% confidence intervals.
Confidence intervals for effects estimates based on linear models are narrower compared to confidence intervals for effect estimates based on quadratic models because including a quadratic term decreases accuracy of estimates.In the models including only the linear predictor, the effect of age never reached substantiality (p < .005and |β| > 0.10 with CIs excluding ±0.10).

Fig. S12
Preference for parenting intention by a linear and quadratic effect of age controlled for a random intercept for country in the sample of bisexual women Note.Parenting intention includes importance rating for a partner sharing the preference for number of children (H7a) and preference for the level of partner's intention to become a parent (H7b).Blue areas represent 99.5% confidence intervals..007 Note.All analyses included a linear effect of age as a predictor and a random intercept for country.Sample size and number of countries differ for reasons explained in detail in the section "Sample Characteristics" of the main manuscript.For all hypotheses the raw and the standardized beta coefficient for the linear effect of age are displayed.Bold effect size estimates indicate substantial effects (p < .005and |β| > 0.10 with CIs excluding ±0.10).b = raw beta coefficient, β = standardized beta coefficient, CI = 99.5% confidence interval.

Fig. S13
Effect size estimates of linear effects on age-range deemed acceptable and on youngest and oldest age deemed acceptable for main analyses and three robustness analyses in the sample of bisexual women Note.Robustness analyses 1: random intercept for language instead of countries; robustness analyses 2: excluding women with missing answers to the seriousness question (nexcluded = 369); robustness analyses 3: random intercept for language instead of countries and excluding women with missing answers to the seriousness question (nexcluded = 369).Graphs display standardized linear age effects, error bars represent 99.5% confidence intervals.Dotted lines indicate the inference criteria for substantial effects (p < .005and |β| > 0.10 with CIs excluding ±0.10).

Fig. S14
Age-range deemed acceptable and youngest and oldest age deemed acceptable by own age in the sample of bisexual women Note.Graphs display means and 99.5% confidence intervals.

Fig. S1
Fig. S1 Partner preferences by own age controlled for a random intercept for country

Fig. S5
Fig. S5 Effect size estimates of linear age effects on partner preferences for main analyses and three robustness analyses in the sample of lesbian women

Fig. S10
Fig. S10 Effect size estimates of linear age effects on partner preferences for main analyses and three robustness analyses in the sample of bisexual women

Table S3
Means (standard deviations) and Cohen's ds for age, partner preferences, and preference for ideal age-range for heterosexual, lesbian, and bisexual women

Table S4
Main results for age effects on partner preferences including preferences for kindnesssupportiveness, attractiveness, financial security-successfulness, confidence-assertiveness, and education-intelligence in the sample of lesbian women

Table S5
Main results for age effects on parenting intention in the sample of lesbian women

Table S6
Main results for age effects on ideal age-range in the sample of lesbian women

Table S7
Main results for age effects on partner preferences including preferences for kindnesssupportiveness, attractiveness, financial security-successfulness, confidence-assertiveness, and education-intelligence in the sample of bisexual women All analyses included a linear effect of age as a predictor and a random intercept for country.Sample size and number of countries differ for reasons explained in detail in the section "Sample Characteristics" of the main manuscript.For all hypotheses the raw and the standardized beta coefficient for the linear effect of age are displayed.The effect of age never reached substantiality (p < .005and |β| > 0.10 with CIs excluding ±0.10).b = raw beta coefficient, β = standardized beta coefficient, CI = 99.5% confidence interval.

Table S8
Main results for age effects on parenting intention in the sample of bisexual women

Table S9
Main results for age effects on ideal age-range in the sample of bisexual women