When planning UX research you have some goal in mind:
- For generative research it’s usually to find out something about users or customers that you previously did not know
- For evaluative research it’s usually to identify any potential issues in a solution
As part of this goal you write down research objectives that help you achieve that goal. But for many researchers (especially more junior ones) they are missing some key steps:
- How will those research objectives help to reach that goal?
- What assumptions have you made that are necessary for those objectives to reach that goal?
- How does your research (questions, tasks, observations, etc.) help meet those objectives?
- What kind of responses or observations do you need from your participants to meet those objectives?
One approach people use is to write their objectives in the form of research hypothesis. There are a lot of problems when trying to validate a hypothesis with qualitative research and sometimes even with quantitative.
This article focuses largely on qualitative research: interviews, user tests, diary studies, ethnographic research, etc. With qualitative research in mind let’s start by taking a look at a few examples of UX research hypothesis and how they may be problematic.
Example Hypothesis: Users want to be able to filter products by colour
At first it may seem that there are a number of ways to test this hypothesis with qualitative research. For example we might:
- Observe users shopping on sites with and without colour filters and see whether or not they use them
- Ask users who are interested in our products about how narrow down their choices
- Run a diary study where participants document the ways they narrowed down their searches on various stores
- Make a prototype with colour filters and see if participants use them unprompted
These approaches are all effective but they do not and cannot prove or disprove our hypothesis. It’s not that the research methods are ineffective it’s that the hypothesis itself is poorly expressed.
The first problem is that there are hidden assumptions made by this hypothesis. Presumably we would be doing this research to decide between a choice of possible filters we could implement. But there’s no obvious link between users wanting to filter by colour and a benefit from us implementing a colour filter. Users may say they want it but how will that actually benefit their experience?
The second problem with this hypothesis is that we’re asking a question about “users” in general. How many users would have to want colour filters before we could say that this hypothesis is true?
Example Hypothesis: Adding a colour filter would make it easier for users to find the right products
This is an obvious improvement to the first example but it still has problems. We could of course identify further assumptions but that will be true of pretty much any hypothesis. The problem again comes from speaking about users in general.
Perhaps if we add the ability to filter by colour it might make the possible filters crowded and make it more difficult for users who don’t need colour to find the filter that they do need. Perhaps there is a sample bias in our research participants that does not apply broadly to our user base.
It is difficult (though not impossible) to design research that could prove or disprove this hypothesis. Any such research would have to be quantitative in nature. And we would have to spend time mapping out what it means for something to be “easier” or what “the right products” are.
Example Hypothesis: Travelers book flights before they book their hotels
The problem with this hypothesis should now be obvious: what would it actually mean for this hypothesis to be proved or disproved? What portion of travelers would need to book their flights first for us to consider this true?
Example Hypothesis: Most users who come to our app know where and when they want to fly
This hypothesis is better because it talks about “most users” rather than users in general. “Most” would need to be better defined but at least this hypothesis is possible to prove or disprove.
We could address this hypothesis with quantitative research. If we found out that it was true we could focus our design around the primary use case or do further research about how to attract users at different stages of their journey.
However there is no clear way to prove or disprove this hypothesis with qualitative research. If the app has a million users and 15/20 research participants tell you that this is true would your findings generalise to the entire user base? The margin of error on that finding is 20-25%, meaning that the true results could be closer to 50% or even 100% depending on how unlucky you are with your sample.
Example Hypothesis: Customers want their bank to help them build better savings habits
There are many things wrong with this hypothesis but we will focus on the hidden assumptions and the links to design decisions. Two big assumptions are that (1) it’s possible to find out what research participants want and (2) people’s wants should dictate what features or services to provide.
One of the biggest problem with using hypotheses is that they set the wrong expectations about what your research results are telling you. In Thinking, Fast and Slow, Daniel Kahneman points out that:
- “extreme outcomes (both high and low) are more likely to be found in small than in large samples”
- “the prominence of causal intuitions is a recurrent theme in this book because people are prone to apply causal thinking inappropriately, to situations that require statistical reasoning”
- “when people believe a conclusion is true, they are also very likely to believe arguments that appear to support it, even when these arguments are unsound”
Using a research hypothesis primes us to think that we have found some fundamental truth about user behaviour from our qualitative research. This leads to overconfidence about what the research is saying and to poor quality research that could simply have been skipped in exchange for simply making assumption. To once again quote Kahneman: “you do not believe that these results apply to you because they correspond to nothing in your subjective experience”.
We can fix these problems by instead putting our focus on research objectives. We pay attention to the reason that we are doing the research and work to understand if the results we get could help us with our objectives.
This does not get us off the hook however because we can still create poor research objectives.
Let’s look back at one of our prior hypothesis examples and try to find effective research objectives instead.
Example objectives: deciding on filters
In thinking about the colour filter we might imagine that this fits into a larger project where we are trying to decide what filters we should implement. This is decidedly different research to trying to decide what order to implement filters in or understand how they should work. In this case perhaps we have limited resources and just want to decide what to implement first.
A good approach would be quantitative research designed to produce some sort of ranking. But we should not dismiss qualitative research for this particular project – provided our assumptions are well defined.
Let’s consider this research objective: Understand how users might map their needs against the products that we offer. There are three key aspects to this objective:
- “Understand” is a common form of research objective and is a way that qualitative research can discover things that we cannot find with quant. If we don’t yet understand some user attitude or behaviour we cannot quantify it. By focusing our objective on understanding we are looking at uncovering unknowns.
- By using the word “might” we are not definitively stating that our research will reveal all of the ways that users think about their needs.
- Our focus is on understanding the users’ mental models. Then we are not designing for what users say that they want and we aren’t even designing for existing behaviour. Instead we are designing for some underlying need.
The next step is to look at the assumptions that we are making. One assumption is that mental models are roughly the same between most people. So even though different users may have different problems that for the most part people tend to think about solving problems with the same mental machinery. As we do more research we might discover that this assumption is not true and there are distinctly different kinds of behaviours. Perhaps we know what those are in advance and we can recruit our research participants in a way that covers those distinct behaviours.
Another assumption is that if we understand our users’ mental models that we will be able to design a solution that will work for most people. There are of course more assumptions we could map but this is a good start.
Now let’s look at another research objective: Understand why users choose particular filters. Again we are looking to understand something that we did not know before.
Perhaps we have some prior research that tells us what the biggest pain points are that our products solve. If we have an understanding of why certain filters are used we can think about how those motivations fit in with our existing knowledge.
Mapping objectives to our research plan
Our actual research will involve some form of asking questions and/or making observations. It’s important that we don’t simply forget about our research objectives and start writing questions. This leads to completing research and realising that you haven’t captured anything about some specific objective.
An important step is to explicitly write down all the assumptions that we are making in our research and to update those assumptions as we write our questions or instructions. These assumptions will help us frame our research plan and make sure that we are actually learning the things that we think we are learning. Consider even high level assumptions such as: a solution we design with these insights will lead to a better experience, or that a better experience is necessarily better for the user.
Once we have our main assumptions defined the next step is to break our research objective down further.
Breaking down our objectives
The best way to consider this breakdown is to think about what things we could learn that would contribute to meeting our research objective. Let’s consider one of the previous examples: Understand how users might map their needs against the products that we offer
We may have an assumption that users do in fact have some mental representation of their needs that align with the products they might purchase. An aspect of this research objective is to understand whether or not this true. So two sub-objectives may be to (1) understand why users actually buy these sorts of products (if at all), and (2) understand how users go about choosing which product to buy.
Next we might want to understand what our users needs actually are or if we already have research about this understand which particular needs apply to our research participants and why.
And finally we would want to understand what factors go into addressing a particular need. We may leave this open ended or even show participants attributes of the products and ask which ones address those needs and why.
Once we have a list of sub-objectives we could continue to drill down until we feel we’ve exhausted all the nuances. If we’re happy with our objectives the next step is to think about what responses (or observations) we would need in order to answer those objectives.
It’s still important that we ask open ended questions and see what our participants say unprompted. But we also don’t want our research to be so open that we never actually make any progress on our research objectives.
Reviewing our objectives and pilot studies
At the end it’s important to review every task, question, scenario, etc. and seeing which research objectives are being addressed. This is vital to make sure that your planning is worthwhile and that you haven’t missed anything.
If there’s time it’s also useful to run a pilot study and analyse the responses to see if they help to address your objectives.
It should be easy to see why research hypothesis are not suitable for most qualitative research. While it is possible to create suitable hypothesis it is more often than not going to lead to poor quality research. This is because hypothesis create the impression that qualitative research can find things that generalise to the entire user base. In general this is not true for the sample sizes typically used for qualitative research and also generally not the reason that we do qualitative research in the first place.
Instead we should focus on producing effective research objectives and making sure every part of our research plan maps to a suitable objective.