To be included in the analysis, a user must have played at least one game in each of the fiveLumosity “Brain Areas”: Speed, Attention, Flexibility, Memory and Problem Solving. The user alsomust have provided their date of birth and gender, between 15 and 85 years of age, and must belocated in the United States according to our geolocation procedure. These inclusion criteriaresulted in an initial cohort of 3,385,648 users.
Geolocation was estimated based on users’ recorded IP address at the time they last logged intoLumosity via the freely availableGeoLiteCity database. This database provides estimates of anumber of different geographic variables, including and estimated city and state and a DesignatedMarketing Area (DMA) code.
Game scores were separately normalized using an inverse percentile rank normalization procedure.Each score was transformed based on its percentile to the corresponding place on a Gaussiandistribution (
= 15). As performance on the exercises is known to decline with age, scores were separately normalized for each age year within each game. By controlling for the effect of agein the population, this procedure avoids biasing the results due to differences in the agedistribution of different cities. Gender differences were then controlled via separate linearregression models for each cognitive area predicting the age-normed score from the user’s gender,resulting in a final scaled score.The overall index of performance was based on a single inverse rank normalization procedure onthe sum of the resulting scaled scores for each cognitive area.
Determining Statistical Areas
Designated Marketing Areas were determined using the Metro Code output of GeoLiteCity. Codes were mapped to DMA labels using atable provided by Google.
Core-based Statistical Areas (CBSAs) and Combined Statistical Areas (CSAs) are defined by the USOffice of Management and Budget based on US Census data. CBSAs include an urban center of atleast 10,000 people, along with adjacent areas that are socioeconomically tied to the urban center via commuting. CSAs group adjoining CBSAs that are also socioeconomically tied, but at a lowerthreshold.
CBSAs and CSAs were determined using two tables that are publicly available on the US Census website. First, the city and state output by GeoLiteCity was mapped to a corresponding county using thenational place file. The county was then tied to a CBSA and CSA (where applicable) using