Within this for each and every site, i obtained step one,one hundred thousand pages regarding heterosexual boys and step one,one hundred thousand profiles regarding heterosexual girls. Pages look for pages through geographical area, decades, and you will gender strain.
To make sure a geographic dispersion off users, i chosen equivalent quantities of pages out-of five biggest towns along with metropolitan, suburban, and you can outlying elements: Los angeles, Denver, Chicago, Atlanta, and you may New york city. I at random selected zip codes of all the four elements to look for pages. Inside for each zip code, for every gender, i after that at random selected pages certainly one of five a long time: early more youthful adulthood (old 18–29 ages; letter = 1,000; Meters = , SD = step three.17), late younger adulthood/early midlife (aged 29–44 decades; n = step 1,000; Meters = , SD = 5.61), later midlife (aged 50–64 decades; n = 1,000; M = , SD = step three.99), and you may older adults (aged over 65 age; n = step one,000; M = , SD = cuatro.29). We utilized such stratifications to assure an entire a long time out of matchmaking pages in the sampling. Since older adults classification you are going to need as much as 30 years, we managed decades as a continuing variable rather than since the an effective group adjustable inside the analyses.
Regarding per reputation, i removed: gender, age, ethnicity, as well as the “On the Me personally” otherwise “Within my Terms” 100 % free reaction section. In order to guarantee privacy off reputation editors, i don’t obtain even more group pointers (e.grams., degree, religious choice, income) that’ll act as distinguishing pointers. The fresh new testing experience illustrated when you look at the Additional Appendix A good .
The brand new attempt varied for the ages out of 18 so you can 95 years. A separate t-shot found no difference in imply many years for ladies (Yards = , SD = ) and boys (M = , SD = ). 
The fresh new writeup on ethnicity throughout the sample try 70% White/Caucasian, 11% Black/African american, 7% Hispanic/Latino, 2% Far eastern, and you may ten% combined battle/most other.
Profile articles
We made use of the LIWC software to research the message of pages. This program exercises the new frequency and you can proportions of specific categories of words in this a book file. The fresh new LIWC program measures up for each and every word of a text file having an interior dictionary greater than cuatro,five-hundred conditions allotted to word classes. This research received into the 11 created LIWC classes: first-people just one pronouns, first-people plural pronouns, members of the family, members of the family, performs, end, currency, health, sex, self-confident emotion, and you will negative emotion. Table step one contains example terms and conditions inside the all the LIWC classification (for more information out-of these types of rules, pick LIWC, 2007).
We together with composed a group of terms having elegance not available into the founded LIWC kinds. We followed steps to possess build out-of LIWC categories ( Tausczik & Pennebaker, 2010) because of the generating an intensive variety of words out of dictionaries, thesauruses, forms out of prior browse, and you can conditions from the research group. Following, we chosen twenty-five conditions very representative regarding elegance according to physical appearance during the thesauruses and you will fellow member answers (elizabeth.grams., adorable, good-appearing, good-looking, hot). Brand new elegance category is almost totally different from the fresh intimate category, with just one overlapping word (sexy). Instances toward attractiveness group are utilized in Table 1; on complete listing of terms and conditions regarding appeal category, discover Supplementary Desk step 1 .
Analytic Approach
I very first looked at detailed analytics towards the proportions of terms some one various years found in its matchmaking profiles. I including produced illustrative figures depicting the most common terms.
I upcoming turned to hypothesis research using average minimum squares regression. The outcome parameters in this data was the fresh ratio from terminology installing each of the twelve categories throughout the LIWC analyses. The brand new LIWC groups was the definitely skewed because of the matter off no values (we.e., new member didn’t have fun with one terminology on the group). I ran for each study which have a rectangular-means conversion process (accustomed target low-normality inside the previous studies utilizing the LIWC; DeAndrea, Shaw & Levine, 2010; Hirsh & Peterson, 2009). This new development regarding conclusions was comparable immediately following applying the changes. Having easy translation, findings is actually showed utilizing the untransformed LIWC classification research. The latest separate adjustable is years, treated once the an ongoing changeable. I and additionally provided sex.