Consumer Housing Trends Report: Methodology

• The Consumer Housing Trends Report (CHTR) includes a sample of successful buyers, prospective buyers, sellers, and renters. In 2026, we asked questions to 144,000+ individuals.
• We use opt-in online panels and statistical methodologies to ensure that our results represent US housing consumers – who they are, what they want, and what they do. Mail-based and email-based surveys would miss more recent movers and skew towards older movers.
• We split what used to be one long survey into numerous shorter, targeted surveys (21 surveys in 2026) to reduce respondent fatigue and fit mobile completion.
• Because opt-in panels skew younger, wealthier, and more educated than the U.S. population, we rake each group’s responses to their Census benchmarks (age, income, education, region, housing tenure, etc.) until the sample matches the country.
• We screen out bots, fraud, and professional survey-takers, then stress-test our numbers by running multiple weighting methods and comparing against Zillow's actual market data.

The Consumer Housing Trends Report (CHTR) provides a comprehensive annual snapshot of how adults across the country experience the housing market as buyers, sellers, and renters. To our knowledge, it’s the largest housing consumer-focused survey in the country. To get a sense of its scale: The 2026 annual survey, fielded March through July across 21 separate surveys, captures responses from more than 144,000 individuals at different stages of their housing journeys. 

This page describes our approach to measuring consumer housing behaviors and ensuring the resulting estimates accurately represent the U.S. population: how we build our sample, how we “weight” it to reflect the U.S. population, and what quality checks we run to make sure the numbers hold up.

How we build our sample

We recruit through online opt-in panels (what we researchers often call "non-probability samples"), rather than the random-digit-dial or address-based methods that used to be the gold standard. Changes in how people respond to surveys and the time demands of a move make this the best source of housing information.

Response rates for traditional methods have, in a word, collapsed. People don’t want to answer a cold call from an unknown number, and younger, more mobile people don’t always check, let alone respond to, their mail. As many housing consumers have just moved or are about to, we’re capturing people at an especially busy, fraught time (34% of buyers report having cried, after all). 

If that wasn’t already enough of a headwind for traditional sampling: the people we most want to hear from are a much smaller share of the US population than many other surveys target. For example, even though they represent millions of people, only 3-4% of U.S. households buy a home in any given year, and fewer still sell. Screening a random sample down to recent buyers or sellers would take an enormous, and enormously expensive, starting pool. And we’d most likely only be able to identify older movers or people who moved a year ago, missing younger movers and those who moved in the last few months.

Opt-in panels solve for speed, cost, and reach into these rare groups. But they also come at the cost of some self-selection (you have to go to a website to sign up to be part of one of these panels). We think that trade off is the right one for CHTR: we'd rather have timely, current-year data on the housing market than wait years for a slower probability-based approach to catch up. And there are some statistical methods (covered in more detail below) that mean well-designed and statistics-aided non-probability surveys tend to track probability benchmarks pretty closely.

CHTR covers multiple consumer groups, each defined by their relationship to the housing market right now:

• Successful buyers bought a home in the past year, including new construction buyers.
• Prospective buyers are actively shopping or plan to buy within the year.
• Sellers sold their previous home in the past six months. We only count existing-home sales here, not new construction.
• Renters split into recent movers, who relocated in the past year, and tenured renters, who've stayed put.
• A general population sample rounds things out, giving us a benchmark for households not currently in the middle of a housing transition.

Everyone we survey is a household decision-maker. That is, the head of household, a spouse or partner, or, in the case of renters, a roommate. Basically, any adult who has a real say in housing choices. We ask it this way on purpose: housing decisions are rarely made by one person in isolation, and surveying only "heads of household" tends to miss younger adults, unmarried partners, and others who are very much part of the decision.

How we ask our questions

Despite its name, the Consumer Housing Trends Report is not a single report. Nor is it a single survey. In fact, it’s over a dozen surveys. (In 2026, the total was 21 surveys.) We do it this way for a few different reasons.

• Doesn’t tire respondents. Older versions of CHTR surveys, like many other surveys, were 20-25 minutes, that’s a long time to ask detailed questions about actions someone took or things someone considered across a multimonth journey. Each survey focuses on a smaller part of the homefinding experience to target respondents’ attention and to be mindful of their time. Plus, we know respondent fatigue is real.
• Meets people where they are. Every year, more and more respondents use their phone to respond to CHTR. What works on a big screen may not work on a pocket, and when people answer on a phone (bus ride, waiting room) isn’t the same as when they’re on a laptop. Shorter surveys are easier to complete in a mobile session.
• Lets us adapt. Part of what we track is changes in the housing market, but we also know the housing market itself is rapidly changing. Smaller surveys allow us to preserve the core questions in the same order to make sure we’re comparing apples to apples. If a new topic emerges that wasn’t on our radar a few years ago (hello, generative AI as an explicit part of the shopping journey!), we can develop a new survey without affecting the way we prime respondents with other content.

How we make sure the results reflect the US population

Here's the problem with opt-in panels: the people who sign up skew younger, wealthier, more educated, and more online than the country as a whole. They’re also more likely to live in some places more than others. Left alone, that skew shows up in every number we publish. Younger, richer people answering for all housing consumers.

So we correct for it with something called raking, and we do it for each group we survey. Think of raking as a dial we turn on each respondent's weight until the survey, taken as a whole, matches the country, taken as a whole, on age, income, education, etc. (wherever the skew shows up). Technically it's an iterative process (the "dial" gets turned many times, a little each time), but the idea is simple: keep adjusting until the mix looks right.

We (well, statistical software does the heavy lifting) determine the right mix based on a few sources. For demographics (age, race/ethnicity, education, income, region, gender, marital status) we use the U.S. Census Bureau's most recent Current Population Survey Annual Social and Economic Supplement. For housing-specific traits (tenure, household composition, building type, depending on the group) we use the U.S. Census Bureau’s 2024 American Community Survey. We stop adjusting once every one of those categories is within 0.5 percentage points of its target, which usually takes 15–20 passes.

Why these particular dimensions and not others? Two reasons: they actually predict housing behavior, and we have solid population numbers to check them against. That second part matters more than people think. We don't weight on things like personality traits, even though they might correlate with how people shop for a home because we can't measure either one reliably, in the survey or in the population. Raking on shaky data doesn't fix bias.

How we ensure we’re measuring things accurately

We keep the data clean

Opt-in panels don't just attract genuine respondents. They attract bots, fraud rings, and people who treat survey-taking as a side hustle. Before any response makes it into CHTR, it has to clear several checks:

• Bot and fraud detection — flags automated patterns: straight-lining through grids, impossibly fast completion times, IPs tied to known fraud networks.
• Professional respondent screening — catches people who show up across many panels or surveys, or who look like they're treating survey-taking as a job.
• Multi-device detection — stops the same person from completing a survey twice on different devices.
• Logic checks — flag answers that don't add up: a $1 home purchase, an implausible age, a "bought in 2020" answer from someone who just screened in as a 2026 buyer.
• Speed checks — flag anyone finishing in under 40% of the median completion time for a closer look.

A respondent usually has to trip more than one of these to get removed. A single odd answer is often just a typo, a stray finger press, or a misread question.

We test our assumptions

Weighting fixes for differences we can see. It can't fix something hiding in the data we can't. (The jargon here is “unobservables.”)  So, we don't stop at one weighting approach; we run several, built on different assumptions, and see whether they agree. Two examples: We rake to a different set of benchmarks, measured a slightly different way.  We also use multilevel regression with poststratification (MrP), which skips reweighting altogether and instead models the relationship between demographics and answers. When methods with completely different blind spots agree, that's a good sign the number's real. 

We check our numbers against other data

CHTR doesn't exist in a vacuum. We line survey responses up against Zillow's own market data to see if what people are telling us matches what's actually happening in the market. For example, does the median time in a home match transaction time in the way we expect? 

For both the assumption checks and alternative data checks, most estimates land within three points of each other no matter which method we use. Where they don't agree — usually for small subgroups, we use caution in our write-up rather than pretend the number is more solid than it is.