Nick Bloom的动态

By the way another bias in their approach is sampling on the outcome. For example, if you wanted to run a survey to find out voting intentions you would not survey 75% of people who vote Republican and 25% of people who vote Democrat. If you do this you will obviously find Republicans at 75% of vote share. You would survey instead a representative sample of voters. If you look at the McKinsey methodology they sampled on the outcome. They sampled on folks "in-person", "hybrid" and "remote". But they *oversampled in-person* and under sampled the other two groups. So guess what, they found more in person employees :-) It is like a "how to" guide on rigging a survey to give you the results you want.

The real-estate folks in McKinsey & Company report office occupancy is flat since 2022. This matches the data I see. The report (linked below) is excellent. It also provides another response to folks asked if McKinsey claims WFH is over - they don't. Most of their material agrees with the public data that WFH is not ending - it is just this one rogue Quarterly piece using biased survey data that claims otherwise. https://www.mckinsey.com/industries/real-estate/our-insights/the-state-of-demand-for-commercial-real-estate

Here is real data on WFH levels over time - plus an underlying source pack: https://www.dropbox.com/scl/fi/imwoo76z963q88t9es830/AEA_2025.pdf?rlkey=9vvosceuqsax06cesdioe25n8&dl=0

McKinsey research is often terrible..in addition to the recall bias, most of their studies have terrible samples, that do not represent any known population, including this one. The samples you use are, of course, much better. I used to work with them years ago, and the strength of the inferences they often made from bad data like this was routine. And they almost never acknowledged the limitations of their data and or that their inferences are, at best, tentative. And I often sent them peer-reviewed studies that had different findings than their in-house research, which they usually ignored. I once got in big trouble for pointing out at conference for a big industry that the reason that--according to McKinsey's research-- there was only one company in the industry using the strategy that McKinsey was recommending (basically, high prices, distinct offerings, high margins) was that many companies in the past had tried it, and all died or could only survive by returning to a low cost/low margin strategy. In other words, survivor bias explained why the cell was almost empty. THEN an industry veteran chimed in, named five or six dead companies that tried the recommended strategy (including a couple he had been part of).

This data from the U.S. Bureau of Labor Statistics (BLS) shows that the percent of hours teleworked as a percent of hours worked during ~January 2025 was ~16.6%, and is at or near the highest point since this data series started in October 2022. U.S. Bureau of Labor Statistics (BLS) data, Percent of hours teleworked as a percent of hours worked, Series Id: LNU0201B8A1 Source: https://data.bls.gov/timeseries/LNU0201B8A1 Also see: https://www.bls.gov/cps/telework.htm

Organizations are naturally going to suffer from cognitive dissonance around this issue and try to retrospectively shift time and engage in this and other cognitive biases to attempt to alleviate it. The as built environment of metro center office buildings, nominally "high speed" highways and automobiles was put in place for a extant time of several decades ago. Information and communications technology and adoption has advanced to the point in the interim it no longer makes sense for many knowledge workers. ICT evolution has been making a new future for knowledge work -- virtualized and decentralized -- since the PC in the 1980s and the Internet in the 1990s/2000s. But the inertia of the as-build pre-1985 environment delayed the future. Until it suddenly arrived in 2020 with the pandemic and is now fueling Tofflerian future shock. Also playing into this is the definition of employment, wherein the employer determines when, where and how work is performed. A return to office mandate fits well with that. But a better fit for manufacturing or other tangible form of work but not as well with much knowledge work.

Nick Bloom, The intent of the research seems be lost here; it was prompted by RTO announcements in 2024, and we wanted to see if it's working. The headline is NOT a decline in WFH; it is (literally in the title) to focus on practices, not just policy. We did not use a representative sample, made clear in method; we used a point-in-time survey for people working under different policies to examine differences *within* our sample. We asked about policies and practices; NOT where people worked on specific days - that was never the intent. Cautions to avoid inferences that assume a representative sample are helpful; but demeaning the research by focusing on what is clearly not the point of the article, and in particular attacking the intentions of the researchers, might be showing your own biases (attribution bias?). RTO policies are ostensibly to drive results like collaboration, innovation, productivity. In-person interactions can help, but our data suggest RTO policies aren't producing substantially better results because other practices are lacking (across working models). If you have read the full article, I would find it more helpful to comment on the actual substance of what the article was about.

Nick Bloom good post. I’ve also been tracking similar trends to the other public data you cite in job postings advertising remote job opportunities. There was a slight dip from pandemic-era highs but remote demand has held steady in recent years at levels that are significantly elevated compared to pre-pandemic norms (about 5x in many cases). Also a good reminder to critically evaluate any research produced by McKinsey or other consulting firms. A lot of their work is excellent, but there is always the potential for tension between what their research may find and what they are incentivized to say (either due to client politics or the solutions they are selling).

Nick Bloom this is the same way I have seen companies massage their data to change the labels of people’s behaviors. Those were used to justify change in stances on RTO. The chart is misleading to show the jump of mostly in person. The best mythbusting is in actual badge in / foot traffic data over a quarter. We know flexibility isn’t about fixed week but rather over the course of a quarter because that’s how business and life schedules work. Why companies with “badge in” monitoring culture and who misrepresent data end up having people figure it out and badge in to fit the labels. McK has done great work in mythbusting with data generally but this one is off. I am a fan of their this week in charts series.

I have to state upfront that, given McKinsey's role in state capture in South Africa (google it), I do not find them a trustworthy organisation. The problem here is that an organisation is not really a thing. It is a collection of people who choose to group together, and that means that you can have excellent people in an untrustworthy organisation. I think the comments below about a bias towards the funder of the research is valid. There is a fine line between telling the client what they NEED to hear and telling them what they WANT to hear, and given that the actual client work is mostly done by junior consultants who are motivated by promotions, it is perhaps not surprising that they do not want to rock the boat. This coming from a consultant who has lost one or two clients because of his directness, has helped some clients immensely because of directness, and has failed to have the impact he could have had on one or two clients because he did not say what the client needed to hear. This is not the arrogance of "look how smart I am" that comes with a big consulting brand. Simply stating that it is difficult to call the line on intellectual honesty, because not all consulting problems have clear black and white answers.