Inside lifestyle filters: what the checkboxes actually do
By Josie Hubschman · Trends · December 12, 2024
TRR has fourteen lifestyle filters on the search page. We get asked all the time which ones actually matter. The answer is: not all of them equally — and the ones that matter most aren't usually the ones people lead with. Here's the breakdown.
The full list
In case you haven't looked recently, our current set:
- Quiet WFH
- Splits chores
- Social
- Early riser
- Night owl
- Cooks at home
- Eats out mostly
- No pets
- Has dog
- Has cat
- Smoker
- Non-smoker
- Hosts often
- Hosts rarely
We've added, removed, and renamed filters dozens of times since 2017. The current set is the one that survived the most testing against actual roommate outcomes.
The two filters that do most of the work
If you only paid attention to two filters, it would be these:
Sleep schedule. Early riser vs. night owl is the single most predictive filter we have. Two early risers get along fine. Two night owls get along fine. An early riser and a night owl can also work — but only if both check Quiet WFH or both check Hosts rarely. Pure sleep mismatch with no quietude compensation produces the most reported conflicts in our data.
Cooking style. Cooks at home vs. eats out mostly predicts kitchen conflict better than any other single signal. Two home cooks can share a kitchen. Two eat-out people don't fight about it. A home cook and an eat-out person can clash over dishes, fridge space, and counter ownership unless they explicitly negotiate it. We don't filter to force compatibility — but we do surface the mismatch.
These two filters alone explain about 70% of the variance in reported roommate satisfaction at three months. That's not a guess — it's from the surveys we send members at the 30, 90, and 180-day marks.
The filters that look important but aren't
Some filters look load-bearing and aren't:
Social vs. not. People assume this one is huge. It's actually weak. Most people self-rate Social the same way most people self-rate clean: optimistically. The real signal is in scheduling — see Hosts often — not in the self-label.
Smoker vs. non-smoker. Critical when relevant, but rarely the issue. About 92% of TRR members self-identify as non-smokers, and the 8% who smoke usually smoke outside or in designated spaces. Smoker/non-smoker conflict is real but uncommon.
The filters we deliberately left out
We've tested filters we ultimately didn't ship:
- Cleanliness level (1–10 scale). Tested in 2018. The variance in what 7 meant was so wide it made the filter actively worse than useless. Self-reported cleanliness is the single most lied-about field on roommate platforms. We replaced it with Splits chores as a behavioral proxy.
- Religious. Tested in 2019. The data showed it didn't predict roommate outcomes at all — religious and non-religious people lived together fine when other lifestyle filters aligned. We removed it.
- Political alignment. Tested in 2020. Predicted outcomes very weakly, mostly through proxy variables we already track. We didn't ship it.
- Income range. People asked for this constantly. We didn't ship it because it leaked sensitive financial data without improving match outcomes. The proxy you actually want is rent comfort, which is already in our budget slider.
How to read filter overlap on a match card
When you look at a potential roommate, the lifestyle pills on their profile match against your filters. Every match card shows you the percentage of your selected filters that overlap. We don't show the math; we show the result.
A 90% match isn't a promise — it's a starting filter. A 60% match isn't a rejection — it's a heads-up that there are tradeoffs to discuss. Use the overlap as a conversation starter, not an answer.
How filter data feeds back into ranking
Every time someone uses the filter set, signs with a roommate, and rates that roommate three months in, we learn something about which filter combinations predict success. We don't share that data; we don't sell it. But we do feed it back into the ranking algorithm, so the next person searching gets a slightly better default ordering than the last person did.
That's what fourteen filters and seven years of testing have produced: not a perfect predictor, but the closest thing we know how to build. The checkboxes do real work — once you know which ones to trust.
.jpg)