Reading summaries like this requires great skepticism. It's exceptionally difficult in this particular case to avoid directing results to arbitrary conclusions through cohort selection and grouping.
> analyzed IP address data in conjunction with LinkedIn data to cross-reference those working from home with those who formed new businesses. ... a title change and employment change on LinkedIn indicating a shift from being an employee to a founder.
Is this more likely to tell us something about the people and roles selected to work remotely, or an outcome of working remotely? At this scale the influences of each are absolutely inseparable. Do cohorts robustly account for education, experience, skillsets, tenure, etc.? The same values which improve one's ability to start a business strongly overlap with the considerations for employing someone remotely. I'm not saying they're comparing a "remote" cohort including developers to a "not remote" cohort including construction workers, but it's important to confirm.
> analyzed IP address data in conjunction with LinkedIn data to cross-reference those working from home with those who formed new businesses. ... a title change and employment change on LinkedIn indicating a shift from being an employee to a founder.
Is this more likely to tell us something about the people and roles selected to work remotely, or an outcome of working remotely? At this scale the influences of each are absolutely inseparable. Do cohorts robustly account for education, experience, skillsets, tenure, etc.? The same values which improve one's ability to start a business strongly overlap with the considerations for employing someone remotely. I'm not saying they're comparing a "remote" cohort including developers to a "not remote" cohort including construction workers, but it's important to confirm.