Statistical analyses

The aim of this analysis is to study and quantify sex differences in first names’ phonetic composition. According to our predictions, we expect to find in the stressed syllable of male names either back or nasal vowels and voiced consonants, as opposed to front, non-nasal vowels and unvoiced consonants in female names. In order to test these predictions, we aggregated all of the first names spanning over the century, giving only one list of first names (e.g. ‘Marie’ was found in several decades). Only one version of phonetically equivalent names in each sex was collated (e.g. ‘Danielle’ and ‘Daniele’, homophones non-homographs, i.e. names pronounced alike but not written alike). Compound names (e.g. ‘Jean-Marie’, ‘Marie-Pierre’) were discarded as they represent a particular set of names mostly composed of a masculine name joint to a feminine name. Monosyllabic names were also discarded from the analysis because it would preclude comparing the first and last syllable. This resulted in a sample size of 275 female and 197 male popular unique names distributed across the century. A generalized linear model was then used to investigate the existence of sex-biased sound symbolic patterns in French male vs. female names. Because the response variable ‘sex’ was binary, a binomial distribution with a logit link function was specified. The explanatory variables were the articulatory features aforementioned, each repeated for the first and the stressed last syllable:

Finally, post-hoc comparisons (Tukey's range test) with a Bonferroni correction were performed for the vowel's place of articulation in order to assess comparisons between the sexes in each syllable. The general size effect was computed using Cohen's f². A symbolic representation of the regression formula is given in the Supplementary Material (Figure S1).

We assessed if the potential significant sound symbolic patterns found in the previous analysis have evolved or remained constant over time between male and female French first names. Pseudo-replication was allowed but phonetically equivalent, compound and monosyllabic names were still excluded, as the aim was to study temporal variations in both the first and last syllable. This resulted in a sample size of 897 female and 790 male names distributed across all decades. To address the time series nature of the data, we first calculated all autocorrelations and partial correlations between each time lag in order to assess if the frequency of a given phonetic variable is dependent on its previous frequency. Vowel articulation was counted as the number of each type of vowel in each syllable and was centred around 0; with 0 corresponding to central vowels, 1 to front vowels and −1 to back vowels. For vowel nasality, it was counted as the proportion of each vowel type: if values are close to 0, first names contain overall fewer nasal vowels, and conversely, if values are close to 1, they contain more nasal vowels. For voiced and voiceless consonants, the mean number in each syllable was studied. Linear models were then used to describe all of the temporal trends. To study possible non-linear effects of time, we modelled a cubic and quadratic effect of decade. Sex was included as another explanatory variable and was put in interaction with time. Model comparisons using the Akaike Information Criterion were then used to assess the best describing model of the temporal variations. All statistical analyses were performed using the R software (version 3.4.4).