6 ms·

> The researchers examined productivity trends in 43 industries — including chemical manufacturing, retail trade, and accommodation and food services — and assi

> The researchers examined productivity trends in 43 industries — including chemical manufacturing, retail trade, and accommodation and food services — and assigned a “teleworkability” score based on the occupational mix of each industry and the share of jobs that can be done remotely. This does not seem like a very strong study.

williamcotton 3y ago
Why?
- Eridrus 3y ago
  I thought this would be obvious, but making up some assumptions about industry-level adoption of remote work based on the composition of employees and then seeing if those industries broke from their pre-pandemic trends requires companies to mechanically adopt WFH policies regardless of whether they are good or not (somewhat reasonable during the peak of covid, but not really for much long after, and also requires the timing of those shocks to line up) and then also needs that shock to be so big that you can see it at the industry level, separate from the general covid shock, and then also risks finding incorrect impacts because covid unsurprisingly had industry-specific effects. Just as a really obvious example: tech stocks soared due to the pandemic; was this because WFH was good for tech companies, or was this due to other factors (wfh means people need to buy more tech, low interest rates pumped up all stocks, etc)? I obviously have my own opinion here, but this study design is not strong enough to disentangle any of this. This study really is total garbage in the vein of "well, our data sucks, but we want to say something" and should not be taken seriously by proponents or detractors of WFH.
- hugh-avherald 3y ago
  To put it pithily: it uses proxies of both dependent and independent variables, which increases the noise of both, in a study to determine signal-to-noise ratio.
  - viraptor 3y ago
    Which proxies do you think are bad / why? The moment you go more coarse than documenting every single worker's perfect timesheet and working out their direct individual output, you're introducing some kind of proxy. What you wrote applies to every economics statistical analysis so far.
    - ororroro 3y ago
      > The researchers examined productivity trends in 43 industries — including chemical manufacturing, retail trade, and accommodation and food services — and assigned a “teleworkability” score based on the occupational mix of each industry and the share of jobs that can be done remotely. They never actually measured the amount of work hours being done remotely. They never measured the number of employees working remotely. They just had some guy make up "remote-ability" numbers for each sector and then carefully pretended to be doing science. Meanwhile all these sectors have seen unrelated changes to operating environment that have not been controlled for. It's BS on BS on BS. Overall I am not even convinced their remote work metric would correlate with hours worked remotely.
    - Eridrus 3y ago
      Actually, this does not apply to good economics papers. Good economics papers have much stronger ways to identify both the independent and dependent variables using microdata from public and private sources. Getting access logs for whether people logged in from home or the office is really not a crazy stretch. Measuring the dependent variable "productivity" is actually much harder since firm-level productivity (aka earnings) has a whole host of other things that impact it besides worker-level productivity and we all know how hard it is to measure developer productivity.

williamcotton 3y ago

Why?

Eridrus 3y ago
I thought this would be obvious, but making up some assumptions about industry-level adoption of remote work based on the composition of employees and then seeing if those industries broke from their pre-pandemic trends requires companies to mechanically adopt WFH policies regardless of whether they are good or not (somewhat reasonable during the peak of covid, but not really for much long after, and also requires the timing of those shocks to line up) and then also needs that shock to be so big that you can see it at the industry level, separate from the general covid shock, and then also risks finding incorrect impacts because covid unsurprisingly had industry-specific effects. Just as a really obvious example: tech stocks soared due to the pandemic; was this because WFH was good for tech companies, or was this due to other factors (wfh means people need to buy more tech, low interest rates pumped up all stocks, etc)? I obviously have my own opinion here, but this study design is not strong enough to disentangle any of this. This study really is total garbage in the vein of "well, our data sucks, but we want to say something" and should not be taken seriously by proponents or detractors of WFH.
hugh-avherald 3y ago
To put it pithily: it uses proxies of both dependent and independent variables, which increases the noise of both, in a study to determine signal-to-noise ratio.
- viraptor 3y ago
  Which proxies do you think are bad / why? The moment you go more coarse than documenting every single worker's perfect timesheet and working out their direct individual output, you're introducing some kind of proxy. What you wrote applies to every economics statistical analysis so far.
  - ororroro 3y ago
    > The researchers examined productivity trends in 43 industries — including chemical manufacturing, retail trade, and accommodation and food services — and assigned a “teleworkability” score based on the occupational mix of each industry and the share of jobs that can be done remotely. They never actually measured the amount of work hours being done remotely. They never measured the number of employees working remotely. They just had some guy make up "remote-ability" numbers for each sector and then carefully pretended to be doing science. Meanwhile all these sectors have seen unrelated changes to operating environment that have not been controlled for. It's BS on BS on BS. Overall I am not even convinced their remote work metric would correlate with hours worked remotely.
  - Eridrus 3y ago
    Actually, this does not apply to good economics papers. Good economics papers have much stronger ways to identify both the independent and dependent variables using microdata from public and private sources. Getting access logs for whether people logged in from home or the office is really not a crazy stretch. Measuring the dependent variable "productivity" is actually much harder since firm-level productivity (aka earnings) has a whole host of other things that impact it besides worker-level productivity and we all know how hard it is to measure developer productivity.