Our why as a company is obvious: to struggle for privateness, selection, and sustainability for sensible properties, and for each one that lives in a single. However who does stay in them?
In December 2024, we launched the primary Dwelling Assistant survey to seek out out. Our aim was easy: to make Dwelling Assistant extra inclusive and approachable by listening on to the varied neighborhood of people that use it.
Within the spirit of constructing within the open, in the present day we’re thrilled to announce the anonymized outcomes of that survey at the moment are freely accessible. On this put up, we’ll run you thru what the survey coated, why and the way we’re publishing this knowledge, what the information shouldn’t be (learn: identifiable), how we’re utilizing this info to enhance what we do, and alternatives for additional understanding and analysis.
TL;DR
- We ran our first Dwelling Assistant neighborhood survey in December 2024, gathering 8,616 responses on how folks arrange and stay with their sensible properties.
- We now have now printed the anonymized outcomes on Zenodo, after a strict anonymization course of – together with k-anonymity evaluate, GDPR compliance, and reducing 12% of responses as a precaution.
- This knowledge is already shaping our work, from redesigning “areas” in Dwelling Assistant to tailoring the Dwelling Assistant Join ZBT-2 and informing our privateness analysis.
- The survey had limitations (it ran too lengthy, and was English-only), which we’re addressing in future analysis.
- Discover the information your self on Zenodo, or the total set of analysis questions and methodology on our GitHub web page.
Discover the information on Zenodo
A snapshot of the survey
To offer somewhat context, the survey ran for about seven weeks, closing in early February 2025, and gathered 8,616 submissions – providing significant insights into Dwelling Assistant customers and the day-to-day realities of working the platform. Most respondents stated an curiosity in know-how drove them to create a sensible house, usually prompted by shifting into a brand new home, or a renovation. Solely round a 3rd began their sensible house with Dwelling Assistant itself, with the remainder arriving from different platforms. Maybe unsurprisingly, amongst those that selected Dwelling Assistant, essentially the most generally chosen causes had been native management (89.3%), its open supply nature (84.0%), and highly effective customization (81.7%).
Past how folks received began, we wished a fuller image of who they’re and the way they stay. The standard person within the examine ran a considerable setup, managing anyplace between 11 and 200 units, and most of the people (75%) did so in a house they owned, and sometimes throughout two or extra flooring. Over half of the households surveyed had a resident cat or canine, and 1 / 4 of respondents spoke multiple language at house.
We additionally requested about skilled background and neurodivergence, since understanding the vary of experience and desires inside the neighborhood shapes how we talk and design: 78.3% of major customers labored or studied in know-how or engineering, and 15.7% recognized as neurodivergent, in contrast with 68.1% who recognized as neurotypical, and 12.3% who had been not sure.
These are only a handful of the attention-grabbing issues we found through this work, and we’ll point out some others all through this put up. For the total particulars, you’ll find the whole checklist of analysis questions, methodology, and different supporting paperwork on our GitHub web page.
Why publish the dataset?
Sharing the data we now have to assist open supply sensible house analysis is an apparent transfer. A wealthy dataset is a robust device, and one we consider everybody ought to have entry to. Publishing the outcomes helps different researchers by permitting them to construct on our work, and means a better range of considering and creativity will be utilized to the information. It additionally makes analysis extra reliable as others can examine the information behind the examine as an alternative of counting on conclusions, and progress is quicker when researchers don’t should re-collect the identical knowledge from scratch.
A part of our mission is to tell society about open supply, moral options to Huge Tech sensible house units. And a technique we do that’s by conducting analysis that generates information, and sharing findings publicly so everybody can study from them. In doing so we’re aligning our group with the ethos of open science: the strategy to scientific publication that believes all publicly helpful analysis and its knowledge needs to be brazenly accessible for all to verify, construct on, and profit from – not locked behind journal paywalls. That’s why we’ve printed on Zenodo, a trusted and broadly used open analysis repository constructed and run by CERN, a world scientific group and pioneer in open supply, open entry, and open knowledge.
Simply as we’d in any formal analysis setting, we maintain ourselves to this excessive commonplace, guaranteeing our strategies and findings are rigorous, verifiable, reproducible, open, and honest.
Championing FAIR knowledge
To be clear, making knowledge technically accessible on-line isn’t the identical as sharing it pretty. That’s why we comply with the FAIR knowledge framework – a broadly used commonplace that ensures printed knowledge is:
- Findable: the dataset has a everlasting hyperlink (a DOI) so it could actually all the time be situated and cited.
- Accessible: anybody can obtain it, no gatekeeping.
- Interoperable: it’s in an open, commonplace format that any device can learn, with a codebook explaining each column.
- Reusable: it comes with clear documentation and an open license so others know the way they’re allowed to make use of it.
FAIR knowledge is what makes info genuinely helpful to a researcher, as an alternative of simply seen. However earlier than we get into how we’re utilizing this knowledge, let’s take a look at what the information isn’t, and what we did to make sure something shared was according to our ideas.
Our publishing strategy: Privateness first
As a basis, we worth privateness a lot that it’s considered one of our three core ideas. So it ought to come as no shock that we adopted a strict protocol to anonymize all knowledge earlier than publishing: no particular person responses, no accounts, no identifiable folks. Right here’s a full rundown of all of the actions we took to guard person privateness:
We eliminated any info that would level to an individual
Names, technical identifiers, timestamps, free-text feedback that would include private particulars, and delicate info pertaining to well being, incapacity, faith, or different family members (for instance, companions and youngsters) had been all eliminated. In actual fact, we deleted a lot figuring out info that the shared dataset is now not thought of “private knowledge” below the Common Knowledge Safety Regulation (GDPR) – a key European privateness regulation.
We made positive nobody stood out via k-anonymity evaluate
Eradicating apparent identifiers isn’t sufficient. Generally a uncommon mixture of atypical particulars (corresponding to a long-time person with an uncommon setup) may make somebody stand out. That’s why we used k-anonymity evaluate – an information anonymization method that checks whether or not sufficient folks share the identical mixture of traits that no particular person will be picked out from the group. We ran validated assessments that appeared for precisely this. The place wanted, we grouped solutions into broader classes: for instance, describing somebody as utilizing “5–6 protocols” relatively than itemizing their actual protocols. This ensured every respondent blended in with many others relatively than being distinctive.
When unsure, we lower the information
In any case these steps, a small variety of responses nonetheless had uncommon sufficient mixtures that we couldn’t be absolutely assured they’d mix in. Slightly than take any possibilities, we merely eliminated them from the dataset. In the long run we overlooked about 12% of responses (roughly one in eight) purely as a precaution. We might relatively publish much less and be sure than threat the privateness of our customers.
We checked it legally
Our strategy was iteratively reviewed by our authorized advisors, and each determination and step is recorded so it may be independently examined.
We printed supplementary materials
Together with the dataset, we launched extra documentation so our actions will be adopted, together with:
- Anonymization protocol and k-anonymity evaluate.
- README file documenting particulars in regards to the survey and the dataset.
- Anonymization and k-anonymity evaluation scripts.
- Full survey query checklist (though not each column on this checklist is included within the dataset on account of privateness, the survey construction itself is efficacious for different researchers).
- Descriptive statistics, masking response distribution for each closed-ended query within the survey.
At each determination fork, we selected the extra cautious possibility, and have deliberately made all our processes clear. In any case, the entire level of this survey is to study with and from our neighborhood, and for us that doesn’t cease on the final query. Which brings us to what the information revealed, and the way it’s already shaping our work.
Understanding social dynamics within the sensible house
Our survey outcomes confirmed that most individuals who responded stay in a sensible house with a accomplice or household. In that context, we discovered that it’s usually one individual – 94.4% of whom are males – who units up and maintains the sensible house, whereas their companions (87.4% of whom are girls) and any youngsters within the family work together with related units relatively than via Dwelling Assistant itself. In different phrases, one individual tends to configure and preserve the system, whereas everybody else merely lives with it.
This survey knowledge aligns with earlier analysis on social dynamics within the sensible house. For instance, research of multi-user sensible properties have discovered that whoever installs the units tends to finish up with outsized management over how they’re chosen, managed, and glued. In some instances, this could result in discomfort round privateness because of this – each from the know-how itself, and from a way of being noticed by the individual managing it (pointing to the potential for this sort of management to be misused), as research by Christine Geeng and Franziska Roesner and Nils Ehrenberg and Turkka Keinonen have individually discovered.
Line Kryger Aagaard’s work has proven that the sensible house expertise is usually gendered, too. Annika’s personal analysis on cohabitation in sensible properties explored this instantly, what occurs when one individual units up and maintains the know-how whereas the opposite lives with it, and discovering energy imbalances that require {couples} to barter management. It’s a dynamic we’re actively working to revamp for (for instance, see ongoing UX design discussions on “Dwelling Assistant for the entire family”), not simply between companions however throughout all kinds of relationships within the house.
How we’re making use of the information
The survey knowledge has already helped deepen our understanding of the person expertise, and the way we optimize it. For instance, info just like the variety of flooring in a house, its dimension, and whether or not it’s owned or rented knowledgeable design modifications for “areas” in Dwelling Assistant. Equally, when our industrial accomplice Nabu Casa was creating what turned the Dwelling Assistant Join ZBT-2 to attach Zigbee and Thread units, we offered them with insights from this survey so they might tailor the gadget to the wants and precise residing situations of our neighborhood.
On a broader stage, the dataset helped us perceive how the neighborhood pertains to the inspiration’s values of privateness, selection, and sustainability – which in flip helped form how we interpreted knowledge inside our privateness analysis, and the way we developed key organizational positions, corresponding to that specified by our Privateness place paper.
Likewise, studying about how present customers arrive at Dwelling Assistant – whether or not they begin with it as their first platform or migrate to it from elsewhere – helps us perceive the expansion of the person base, and the way the inspiration can higher assist our neighborhood to advance our mission (hold a watch out for an in depth report on this quickly).
After all, as with every scientific analysis, the survey has limitations. It’s necessary to call them, each to be clear and since they open alternatives for future enhancements.
Limitations of the survey
The survey was too lengthy. Lots of you pointed this out to us, and regardless of having timed the survey at about 20 minutes throughout assessments, we had been clearly a bit off. In line with Typeform (the survey platform we used), the common completion time was over 40 minutes! We admire anybody who began the survey, and are particularly grateful for all who caught with it till the top. Having such rigorous, engaged neighborhood members is an effective drawback to have. However it could actually additionally skew outcomes, since solely these with the time and curiosity to complete a protracted survey are represented. Due to that, and the actual fact the information doesn’t seize responses of those that dropped out of the survey – respondents symbolize a dedicated core of Dwelling Assistant, not all customers.
As some neighborhood members identified, the survey being solely in English additionally restricted its attain and presumably its accuracy. Quite a few Dwelling Assistant customers that talk languages aside from English weren’t represented, and the place non-native English audio system did fill out the shape, it’s potential that some questions could not have landed clearly.
Language obstacles, biases, and blown-out survey instances are all issues that needs to be taken into consideration when decoding the information. Analysis gaps are regular in academia – we point out them as a result of we’re conscious of the place we are able to enhance, and to set ourselves up for future analysis that’s as inclusive as potential. However there’s additionally a chance right here… In the event you’re a researcher engaged on sensible house know-how in academia or trade, these gaps is perhaps a helpful place to begin in your personal work. In the event you’d like to debate this analysis or discover future tasks collectively, attain out to us through e-mail or be a part of the UX design or person analysis discussions on GitHub. We’d love to listen to from you.
Future analysis
Lots of you shared nice concepts for extra survey questions, masking the established order of sensible properties, future plans, objectives, and desires, different methods of interacting with the neighborhood, and present friction factors inside Dwelling Assistant.
The beneficiant suggestions on the content material, framing, and even the styling of the survey from neighborhood members was drastically appreciated, and factors to one of many tensions we bumped into whereas designing this survey within the first place: there’s an excessive amount of to cowl! With so many attention-grabbing matters to discover, narrowing the questions down was a problem, and an enormous a part of why the survey ended up so long as it was.
However that’s additionally the fantastic thing about person analysis: there’s all the time a lot to study! For that cause, we’re planning one other neighborhood survey, so keep tuned!
Conserving the dialog going
At its core, this survey was about listening: changing our guesswork together with your info and opinions, so we are able to construct Dwelling Assistant for individuals who truly stay in a sensible house, not simply who we assume does. Publishing this dataset brazenly is one other approach of placing our ideas into observe – letting anybody dig into what we’ve discovered for the good thing about everybody.
None of this may exist with out the hundreds of you who took the time to reply, and caught with a survey that ran longer than we meant. Thanks. Right here’s to extra open, sincere conversations with the neighborhood that makes this all potential.
