Within for each and every website, i amassed step one,100 users of heterosexual people and you can 1,one hundred thousand users regarding heterosexual ladies. Pages seek out profiles through geographical place, ages, and you will intercourse filter systems.
To be sure a geographic dispersion off pages, i picked equivalent numbers of pages of four major towns plus metropolitan, residential district, and you can rural section: Los angeles, Denver, Chicago, Atlanta, and you can New york city. I at random picked zip rules from each of the four portion to search for profiles. Contained in this for each area code, per intercourse, we next at random chosen users certainly one of four age groups: very early more youthful adulthood (aged 18–30 many years; letter = step 1,000; Meters = , SD = step three.17), late young adulthood/very early midlife (old 31–44 ages; letter = 1,000; M = , SD = 5.61), later midlife (aged 50–64 age; letter = 1,000; M = , SD = step three.99), and the elderly (aged more than 65 age; letter = 1,000; Yards = , SD = 4.29). I used these stratifications in order to guarantee a full a long time of matchmaking users when you look at the testing. Because the older adults class you will incorporate to 3 decades, i managed age because the a continuing varying instead of since good collection varying in analyses.
Out of for every profile, we removed: gender, many years, ethnicity, as well as the “On Me” or “In my own Conditions” 100 % free impulse area. In order to guarantee anonymity of reputation publishers, we did not obtain even more group information (age.g., degree, religious choice, income) that will act as pinpointing information. The brand new testing method is portrayed inside Second Appendix A great .
The new shot varied in ages out-of 18 so you’re able to 95 years. Another t-sample revealed zero difference in indicate years for women (Yards = , SD = ) and males (Meters = , SD = ). The brand new summary of ethnicity on shot was 70% White/Caucasian, 11% Black/Dark colored, 7% Hispanic/Latino, 2% Far eastern, and you will 10% combined battle/almost every other.
Character content
I used the LIWC software to research the content of profiles. This software works out the brand new volume and you can size of certain categories of conditions contained in this a book document. The fresh LIWC system compares for each word of a book document having an interior dictionary greater than 4,500 terms and conditions allotted to keyword groups. This study drew to your 11 established LIWC categories: first-people singular pronouns, first-individual plural pronouns, nearest and dearest, family unit members, functions, completion, currency, health, sex, positive feelings, and you can bad emotion. Table step one consists of analogy terms during the all the LIWC class (to find out more out-of such rules, see LIWC, 2007).
We as well as composed a group of conditions for appeal not available for the depending LIWC categories. I used procedures to 
possess build out of LIWC categories ( Tausczik & Pennebaker, 2010) of the generating a thorough list of terms regarding dictionaries, thesauruses, forms from past browse, and you will terms and conditions generated by the research cluster. Upcoming, i picked twenty five terms and conditions very user out-of elegance considering appearance during the thesauruses and participant answers (elizabeth.grams., sexy, good-lookin, good-looking, hot). New attractiveness group is almost completely not the same as new intimate group, with only one overlapping keyword (sexy). Instances on attractiveness class are used in Dining table 1; on done variety of terms from the attractiveness classification, pick Supplementary Dining table step 1 .
Analytical Means
I very first checked descriptive analytics on the proportions of terms somebody of different ages used in their relationship pages. We and generated illustrative rates portraying the most famous terms and conditions.
I then looked to theory testing using average least squares regression. The results variables within studies were the fresh proportion from words suitable each of the several classes in the LIWC analyses. Brand new LIWC kinds was basically most of the positively skewed considering the count regarding zero values (i.e., participant don’t play with people conditions in the group). We went for each analysis having a square-root conversion (regularly target non-normality in early in the day training making use of the LIWC; DeAndrea, Shaw & Levine, 2010; Hirsh & Peterson, 2009). The fresh new development regarding conclusions was similar after applying the changes. To possess easy translation, conclusions was displayed utilizing the untransformed LIWC class data. New separate changeable was many years, managed as a continuing varying. I and incorporated gender.