Multimedia Appendix 1: Supplemental Methods 

 

Supplemental Methods 1:  Detailed Data Extraction Description 

 

Several factors pertaining to obtaining Twitter Premium API data contribute to the discrepancy 

between the 12,337 novel English geotagged tweets and the subsample sample of 11,935 novel 

English geotagged tweets that we use in this study.  

 

The Twitter Premium API has a maximum retrieval limit of 500 tweets per data request and each 

request costs money. Therefore, we had to optimize our requests. We decide to download tweets 

based on non-overlapping times in which they originated. This process led to the loss of some 

tweets during periods with more than 500 Tweets. For example, if we requested Tweet between 

8pm-9pm on Oct. 15 and there were over 500 Tweets, say 637, we recovered only the first 500.  

The remaining 137 Tweets would have to be recovered in a separate request. Therefore, we 

examined the data to try to ascertain the time of last tweet downloaded to then send a new 

request with non-overlap times to fill in missing tweets. However, this process was inherently 

error prone. In some cases, we lost a few minutes and in others we overlapped in time. This led 

to both some duplicate Tweets and brief periods with missing data.   

 

After data cleaning we had 11,935 non-duplicates tweets but based on the Twitter API counts 

there were 12,337 that should have met our search criteria. However, the counts data depend on 

the number of non-deleted tweets at the time of query. Twitter support documentation states, 

“The counts delivered through this endpoint reflect the number of Tweets that occurred and do 

not reflect any later compliance events (deletions, scrub geos). Some Tweets counted may not be 

available via data endpoint due to user compliance actions.” We queried the counts before 

retrieving the data, so in addition to having brief periods with missing data, some small number 

of tweets may have been deleted by the users. 
 

  



Supplemental Methods 2: Description of the Demographics Pro Algorithm  
 

Demographics Pro is a commercial service used to assess demographics and other characteristics of social 

media populations. We used their services to infer gender, age and race/ethnicity of the posters of 

revealed experiences.  We provided Twitter handles, or screen names, of posters categorized as having 

revealed an experience of sexual assault/abuse and early life experience of sexual assault/abuse. 

Demographics Pro then provided us with a distribution of gender, age and race/ethnicity for the handles 

we shared and similar data for a random sample of 20,000 Twitter users in the US as a comparison.  

 

Demographics Pro uses a series of proprietary machine learning algorithms to estimate or infer likely 

demographic characteristics of Twitter handles based on Twitter behavior/usage. Their predictions rely on 

signals from networks, consumption, and language use.  They have created and validated their 

methodology against 300 million Twitter users.  They require 95% confidence to make an estimate on a 

single demographic characteristic of a handle. Importantly, they do not return individual level data to 

researchers for us to make independent assessments of their algorithms, rather they return aggregates on 

the sample. This is to ensure privacy for Twitter users.   

 

See http://www.demographicspro.com/views/demographics_aboutmeth.shtml for more details.  

 

http://www.demographicspro.com/views/demographics_aboutmeth.shtml