Using Regression to Understand Users
A challenge that has always plagued the web industry is understanding users. Whether it’s their habits, assumptions, interests, tendencies, or expectations, it all affects how any user will interact with and respond to a website. Since no group of users or customers is the same, it takes more than just a general knowledge of what users want in order to design something effective.
Traditional UX research practices include user interviews, focus groups, and usability testing to answer some of these questions. These tools are great for being able to dive deeper into questions, asking why, probing where needed, and developing empathy for how certain people think. However, there’s a potential gap if we rely solely on these tools. These exercises are anecdotal, and thus only tell a piece of the story.
What if the five users you interviewed are not representative of the entire customer population? There needs to be some way to verify conclusions across a larger sample, substantiating your findings with real, hard data.
There are a few ways to collect data across a larger sample including Google Analytics, ClickTale, client records, and surveys. Surveys, specifically, provide a great way to answer particular questions that could provide invaluable insights beyond past user interaction. The output provides not only statistical and basic conclusions like 60% of the customer base is male, but also answers questions about how variables might relate to and affect each other. This can be done using regression analysis.
Regression analysis in a nutshell is a mathematical method of determining what independent or causal variables change or contribute to a certain dependent or effect variable. Consider the following question:
How does gender, ethnicity, and education level affect how often a customer makes online purchases?
In this case, gender, ethnicity, and education would be considered independent variables and frequency of online purchases the dependent variable. If you’re interested in a more thorough explanation, feel free.
To provide a brief road map, the list below is the process that we normally follow when conducting research utilizing regression techniques:
- Develop hypotheses
- Design survey questions
- Send to a representative sample
- Collect and clean data
- Run regressions to test hypotheses
- Test regressions
- Rerun regressions based on tests
- Report findings
The last step is obviously the most fun where all of your hard work pays off in meaningful insights about your users or customers. Based on these conclusions, personas can be developed, interactions designed, features prioritized, and entire strategies built with support from real data.
To provide some context, we recently engaged in the above process for an ongoing project with a prominent brand. We sent our designed survey to all of their online customers, receiving 4,000 responses. Analyzing approximately 55 variables, we reached some interesting and compelling conclusions, such as:
Individuals who do more online product and price comparisons are more likely to agree that ________ is expensive.
Males visit _________.com more frequently than do females.
Younger customers are more likely to shop around for the best product and price when shopping online.
Listeners of Classic Oldies, Country, Gospel, Metal, and Rock music genres are more likely to use a CD player to listen to music.
Males are more likely to purchase digital music than are females.
Conclusions like these are invaluable to our research and strategic recommendations as we design products that affect the entirety of any brand’s customers. While this is still only a piece of the puzzle, it’ll continue to be an illuminating and rewarding process in understanding users and how to design the user-centric systems and platforms that Viget prides itself on.