This appendix offers information about two important components of our research: the survey that we conducted to understand the behavioral shifts caused by the pandemic, and the models that we built to project demand for office, residential, and retail space in several scenarios.

Our survey

In chapter 1, we describe many pandemic-induced behavioral shifts. We deduced some of those shifts from a large international survey that we conducted in October and November 2022 to learn how office workers’ behavior had changed—in ways that could affect real estate—during the pandemic. We surveyed nearly 13,000 full-time office workers, all of them at least 18 years old, in six countries: China (about 2,600 people), France (1,400), Germany (1,900), Japan (1,700), the United Kingdom (2,300), and the United States (3,200).

We ensured that the initial set of respondents who received the survey had the same distribution of age and sex as their country did. In the United States, where we had more information about potential respondents, we were also able to account for income, ethnicity, and region (that is, the Northeast, Midwest, South, and West).

We considered responses only from people who met certain requirements, such as being full-time office workers or remote workers performing office-related duties. That group was called the target audience.

Then, in order to increase the number of responses, we sent the survey to a larger set of respondents, one whose demographic characteristics matched those of the target audience. We then weighted respondents from each country to reflect the country’s distribution of age and sex within the population of office workers (a distribution that we determined by examining the target audience’s responses). Once again, we were able to weight US respondents by income, ethnicity, and region as well. Where we present global averages, we simply averaged results from the countries we studied rather than weighting those results by population.

We intentionally overrepresented certain groups. First, we overrepresented respondents from 17 cities of particular interest to make sure that we had enough information about those cities. The cities were Atlanta (where we surveyed about 480 people), Beijing (540), Berlin (400), Chicago (460), Chongqing (70), Hong Kong (500), Houston (440), London (840), Los Angeles (440), Munich (550), New York City (440), Osaka (490), Paris (730), San Francisco (410), Shanghai (560), Shenzhen (490), and Tokyo (560).¹Those numbers do not add up to 13,000 because we also surveyed people from other cities. Second, we overrepresented people who had moved after the pandemic began to make sure that we had enough information about migration patterns.

The questions that we asked covered the following areas:

• Respondents’ personal characteristics (for example, “What is the highest degree or level of school you have completed?”).
• Employment information (for example, “What is your typical commute time to the office, one-way?”).
• Office attendance and factors motivating working from home (for example, “On average, how many days of the week did you work in the office pre-COVID [late 2019]?” and “Why do you typically choose to work in the office, on days you do not work from home?”).
• Trade-offs related to working from home (for example, “How much of your compensation [as a percentage of your pretax salary and bonus] would you be willing to forgo to be able to work from home for your ideal number of days per week versus being in the office every day?”).
• Space needs (for example, “For the time that you are in the office today, how has your allocation of time among types of spaces changed relative to how you utilized spaces before the pandemic [that is, late 2019]?”).
• Retail spending (for example, “How has your share of retail expenditure at the following locations shifted today [fall 2022] versus pre-COVID [late 2019]?”).
• Most recent move and motivating factors (for example, “In what ways did the pandemic influence your move?”).
• Impact of move on preferences and persistence (for example, “In what ways has working from home changed your preferences around where you live and what kind of space you need?”).

Finally, we ensured that each part of our analysis had a sufficiently large sample. For example, every industry shown in Exhibit E1 had at least 300 respondents.

Our models

In chapter 2, we offer projections of demand for real estate in 2030. Those projections were generated by three interrelated models we built, one apiece for office, residential, and retail space. We used each model to project demand in three scenarios—called reversion, moderate, and severe—in which the pandemic’s effects were increasingly strong. The factors that each model took into account (some of which were drawn from another model’s output) and the ways the scenarios differ from one another are described below.

Office space model

Our model for projecting demand for office space considered the following four factors:

• Office employment, defined as the total number of office workers in an urban core.
• Office attendance, defined as employees’ current office attendance as a percentage of their office attendance before the pandemic. For example, if an employee went to the office five days a week before the pandemic but now goes to the office three days a week, that employee’s office attendance is 60 percent.
• Employee coordination, defined as the maximum share of workers in the office at a given time that employers need to provide space for. This factor is important because office attendance figures alone may overstate the degree to which demand for office space has fallen. If all employees go to the office only once a week, but all are present on Tuesday, demand will be no smaller than it was when they worked in the office daily. By contrast, if all employees go to the office three days a week, but the company coordinates attendance so that no more than 60 percent of employees are present on any day, demand for space is reduced by 40 percent. Our model assumes that employers will enforce coordination to capture half of the space savings from reduced office attendance. For example, if attendance decreases by 40 percent, the model assumes that demand for square footage will drop by 20 percent.
• Space per seat, defined as the total usable square footage allocated to each office worker. Space per seat was declining even before the pandemic.

We varied the values for the first two of those four factors to estimate demand in three scenarios.

• In the reversion scenario, we essentially assume that the world will revert to its condition before the pandemic. Office attendance returns to its prepandemic level by 2022 and stays at that level through 2030. Office employment growth is based on the historic rate of population growth as well as shifts in labor force participation rates and in the share of office workers. Employee coordination and space per seat are held constant at their current levels.
• In the moderate scenario, the pandemic’s effects on population, and therefore on office employment and office attendance, are stronger. (The effects on population are drawn from the moderate scenario in our model for residential space, which is described below.) By 2025, office attendance is higher than it is now but still lower than it was before the pandemic, and that partial recovery continues indefinitely. Employee coordination and space per seat are again held constant at their current levels.
• In the severe scenario, the pandemic’s effects are long-lasting and reduce population growth, so the effects on office employment and office attendance are even stronger. (The effects on population are drawn from the severe scenario in our model for residential space.) Office attendance remains at current levels indefinitely. Employee coordination and space per seat are again held constant at their current levels.

Residential space model

Our model for projecting demand for residential space considered the following three factors:

• Population, defined as the current number of residents in an area. We modeled population separately for the urban core and suburbs of each city (see Box E1, “How we define cities,” in the executive summary). During the pandemic, increased out-migration from urban cores to suburbs disrupted population growth.
• Average household size, defined as the average number of members in each household. We assumed that through 2030, average household size would change at its historical rate from 2010 to 2020.
• Average size of home, defined as the average number of rooms per home in the United States and the average number of square feet per home in the other countries we studied. Here too, we assumed that through 2030, average household size would grow at its historical rate from 2010 to 2020.

We varied the values for the first of those three factors to estimate demand in three scenarios.

• In the reversion scenario, people who moved during the pandemic return to their original residences in urban cores or suburbs by 2025, driven by a full return to prepandemic rates of office attendance. Because of their return, excess migration, which we define as out-migration from urban cores during the pandemic that exceeded the 2015–19 average, falls below zero. After 2025, population growth returns fully to its prepandemic rate (that is, the annual rate from 2010 to 2020).
• In the moderate scenario, people who moved during the pandemic do not move back, because hybrid work is now the standard working model. Excess migration continues, but it is small and gradually falls to zero, and population growth in most cities returns to its prepandemic rate by 2030.
• In the severe scenario, people who moved during the pandemic do not move back. Excess migration continues, is high, and does not fall to zero until after 2030. As a result, population growth in most of the cities we studied is much lower than it was before the pandemic.

Retail space model

Our model for projecting demand for retail space considered the following six factors:

• Population, defined as the current number of residents in an area. We modeled population separately for the urban core and suburbs of each city.
• Impact of remote work on spending, defined as the change in retail spending in physical stores resulting from changes in office attendance as commuters come to workplaces less often and spend less money in stores near them.
• E-commerce penetration, defined as online spending as a share of total retail spending. In the early years of the pandemic, e-commerce penetration grew more quickly than it had historically because of quarantines, social distancing, and increased remote work.
• Sales productivity, defined as in-store retail sales per square foot of occupied retail space. For example, if a store with 100 square feet of space generated $1,000 in annual sales, sales productivity would be $10 per square foot.
• In-store retail sales, defined as annual retail sales that occur in a physical store within a metropolitan area. To determine the growth of in-store retail sales, we began with the projected rate of general consumer spending at the city level through 2030 and adjusted it for population change, the impact of remote work on spending, and e-commerce penetration.
• Commuters, defined as the number of people who commute to the urban core for work. We assumed that the ratio of commuters (but not the number of commutes) to the population of the urban core would remain constant at its 2019 value through 2030.

We varied the values for the first four of those six factors to estimate demand in three scenarios.

• In the reversion scenario, as a result of employees’ full return to the office, office commuters’ retail spending near the office returns to prepandemic levels by 2025. The population assumptions are drawn from the reversion scenario in our residential space model, and the office attendance assumptions that inform the impact of remote work on spending are pulled from the reversion scenario in our office space model. E-commerce penetration returns to its prepandemic trend line by 2025. Sales productivity remains at its 2019 level.
• In the moderate scenario, employees engage in only a partial return to the office. The population assumptions are drawn from the moderate scenario in our residential space model, and the office attendance assumptions that inform the impact of remote work on spending are pulled from the moderate scenario in our office space model. E-commerce penetration again returns to its prepandemic trend line by 2025. Sales productivity remains constant at its 2019 level in all cities except Beijing and Shanghai, where recent growth gives us reason to expect a future increase.
• In the severe scenario, retail spending remains at its current level. The population assumptions are drawn from the severe scenario in our residential space model, and the office attendance assumptions that inform the impact of remote work on spending are pulled from the severe scenario in our office space model. Pandemic-induced increases in e-commerce penetration continue, except in cities where e-commerce penetration has already returned to prepandemic rates of growth. Sales productivity grows slightly in all cities.