Cycling women (not on hormonal birth control)

Women on hormonal birth control

After publication, we issued a number of corrections and clarifications. Details can be found in the following places.

This online supplementary documents our code and results. It is part of the online supplement on the Open Science Framework, where you will also find the text of our preregistration, the files necessary to reproduce the study structure and items in formr.org, synthetic data, and with access rights, the anonymised real data.

Below follows the table of contents for this website. Every link is followed by a short explanation.

We performed a power analysis to compare our diary design (with many days per woman) to more common 2-day and 4-day designs. To see how exactly the simulations were run, confer the cluster script.

We also performed researcher degrees of freedom analyses to better understand some of the research practices that have been discussed as possible reasons for false positive inflation in the literature. To see how exactly the simulations were run, confer the cluster script

Blake et al. re-analysis. Blake et al. (2017) analyzed luteinising hormone data and concluded a very low reliability for day-counting methods. However, their reliability numbers apply to a different set-up than ours, so we re-analyzed their data, finding improved performance for day-counting.

OCMATE data, Jones et al. re-analysis. Jones et al. (2018) provided open data for their hormonal measurements. We used these to compute reliabilites of change in addition to their reported coefficients of variation.

Data wrangling. This document shows all R code we needed to get from the raw data exported from formr to the data used in our analyses. It includes detailed information on fertile window determination, exclusion criteria, scale aggregation, and participant flow.

Descriptives. This document has extensive descriptive information on main demographic variables, but also includes a flow chart of exclusion criteria, the comparison between the naturally cycling group and the hormonal contraceptive user quasi-control group, missingness patterns for the different fertile window predictors, generalizability coefficients (multilevel reliabilities) for the diary items, measurement reactivity examinations, a detailed analysis of contraceptive methods, and the geographic origin of participants.

Pre-registered analyses. This document shows the results of our preregistered tests. In addition to the information in the manuscript, this page also contains marginal effect plots of all analyses, further details of moderator tests, and the detailed results of adjusting for self-esteem. The page is organised according to outcomes. If you click an outcome tab, you can choose from the results with the “narrow window”, the “broad window”, and, where predicted, the moderator results.

Robustness analyses. This page documents the majority of our robustness checks. Please allow some time for the page to load, as these were very extensive. The page is organised by outcome like the preregistered analyses. For each outcome, you can find 1) the summary of the main model, including marginal effect plots, the outcome distribution, and regression diagnostics, 2) continuous curves of the outcome over cycle days, with varying assumptions, 3) the robustness checks, which include alternative predictor specifications, covariate sets, regression assumptions, methodologically relevant moderators like cycle length, etc. 4) where predicted, moderator analyses using the minimal exclusion criteria dataset, including some alternative moderator specifications. This page also includes results for outcomes where we did not predict an effect, and response time, as a potential confound. This table gives descriptions of all the robustness checks that were conducted for all outcomes.

Bayesian item-level ordinal models for extra-pair desire and behaviour). Because the assumption of normality was violated for the extra-pair items, we additionally ran Bayesian models with an ordinal specification. These models also document an attempt to get appropriate standard errors for the cycle curves, and an examination of differences in fertility effects across items and persons.

Bayesian item-level ordinal models (separate by item). This page documents Bayesian ordinal models run separately for each item. These results have to be regarded as exploratory, but may help researchers specify their outcomes and predictions in future confirmatory studies. This page includes a plot and table of effect sizes across items.

Curve plot for paper. Reproducible code for making our curve plot figure in the manuscript.

Compare in- vs. extra-pair desire vs. libido explanations. An in-depth look at the challenges of comparing the effect size for extra-pair desire and in-pair desire, and making conclusions about potential underlying ovulatory changes in undirected sexual desire.

Data anonymisation procedure. Documentation of how we conducted the lossy anonymization of our data, and how we generated synthetic data.

In this document, all Helper functions that are used throughout, are documented. These are the .Rprofile settings

The statistical models for each outcome were documented in a standardised fashion using rmarkdown component files (partials), documented below: