Quantitative research is how you find out how many, how much, and how often. It turns behavior and opinion into numbers you can count, compare and test, which is what makes it the backbone of most product and UX decisions that need to hold up in a room full of stakeholders.
Most people arrive at this topic with one question: how many types of quantitative research are there, and which one do I actually need? The short answer is four, and they form a ladder from simply describing what is happening to proving what caused it. Below you will find the four types summarized up front, then each one explained with a current, concrete example, the three main data-collection techniques, and a method for choosing between them.
The 4 types of quantitative research
- Descriptive research — measures and summarizes what is happening, without touching any variables. Use it when you need a baseline: how many users, how often, how satisfied, how long a task takes.
- Correlational research — measures whether two or more variables move together, and how strongly. Use it when you want to know if a relationship exists (and how to prioritize) but cannot or should not run an experiment.
- Quasi-experimental research — compares groups that were not randomly assigned, usually because a change already happened. Use it when you want cause-and-effect evidence but random assignment is impossible or unethical.
- Experimental research — manipulates one variable at random and measures the effect on another. Use it when you need to prove that a change caused an outcome, not just that the two co-occur.
Descriptive tells you what. Correlational tells you what goes with what. Quasi-experimental and experimental tell you what caused what, with experimental giving you the strongest claim because of randomization (Creswell & Creswell, Research Design, SAGE).
What is quantitative research?
Quantitative research collects numerical data from a sample and uses statistics to draw conclusions about a wider population. Because the output is numbers, results can be aggregated, compared across segments, tracked over time, and tested for statistical significance.
In contrast to qualitative research, quantitative research works best with larger samples and closed-ended measures. Qualitative asks "why?"; quantitative asks "how many?" and "how much?" Nielsen Norman Group frames the same split as a question of what the data lets you conclude: quantitative methods give you magnitude and confidence intervals, qualitative methods give you the reasons behind the number (NN/g, Quantitative User-Research Methodologies).
The trade-off is depth. Numbers are easy to analyze and hard to interpret on their own, which is why most mature research programs pair the two into mixed methods: the survey tells you how many trial users never reach a second session, the interviews tell you why.
Photo by Firmbee.com on Unsplash
Types of quantitative research methods
If your objective is gathering numerical data, the first decision is which of the four designs fits the claim you need to make. A method that works for one team's question will produce an unusable answer for another, so start from the claim and work backwards. For a wider view of where these sit among all the types of research, see the overview post.
The four designs, in order of how much they let you claim:
- Descriptive research
- Correlational research
- Quasi-experimental research
- Experimental research
Descriptive research
Descriptive research measures variables and reports them as they are. It does not manipulate anything and it does not establish causation; it establishes the picture. Frequencies, means, medians, distributions and percentages are its output, and it is often called observational research because you are counting rather than intervening.
2026 product example. A B2B analytics team wants a baseline for onboarding. They pull product analytics for every account created in the last quarter and report: median time to first successful import, the share of accounts that complete each onboarding step, and task-completion rates from an unmoderated benchmark test. Nothing is manipulated. The deliverable is a dashboard and a set of numbers the team can now measure against.
Descriptive studies come in cross-sectional form (one snapshot in time, across a diverse sample) and longitudinal form (the same subjects measured repeatedly, which is how you see trends and drift). Pew Research Center's panel-based surveys are the canonical large-scale example of the longitudinal version (Pew Research Center methods).
Correlational research
Correlational research measures two or more variables and tests whether they move together, without manipulating either. The output is a correlation coefficient and a significance level: strong or weak, positive or negative. It cannot tell you which variable caused the other, or whether a third variable is driving both.
2026 product example. A team suspects that teams who invite a second seat in week one retain better. They pull the seat-invite flag and 90-day retention for every account, and run a correlation. If the relationship is strong, that is a prioritization signal worth acting on and worth testing properly. It is not yet proof that inviting a teammate causes retention: it is entirely possible that already-committed teams do both.
Correlational work is cheap because it usually runs on data you already have, which makes it the natural first pass before you spend an experiment slot. It is also where most misreported research goes wrong, so state explicitly in the report that the finding is associational.
Quasi-experimental research
Quasi-experimental research compares groups that received different treatments, but without random assignment. You lose the guarantee that the groups were equivalent to begin with, so you compensate with design: pre/post measurement, matched comparison groups, difference-in-differences, or an interrupted time series (Trochim, Research Methods Knowledge Base: quasi-experimental design).
2026 product example. A pricing page was redesigned and shipped to everyone at once, so there is no control group. The team compares conversion for the eight weeks before and the eight weeks after, and uses a second, untouched market as a comparison line to absorb seasonality. That is a quasi-experiment. It is weaker than a randomized test because something else may have changed in the same window, but it is a legitimate design and often the only one available for launches, policy changes, migrations and anything you cannot ethically withhold.
Experimental research
Experimental research manipulates an independent variable, assigns participants to conditions at random, and measures the effect on a dependent variable. Randomization is the whole point: it is what makes the groups comparable and lets you attribute the difference to the change.
2026 product example. An online A/B test on a signup flow, with users randomly assigned to the current form or a version with one fewer field, powered to detect the smallest effect the team would act on, and read once at a pre-declared sample size. Multi-arm tests, factorial designs and holdout groups are all variants of the same logic. The discipline that separates a real experiment from a dashboard comparison is deciding the metric, the sample size and the stopping rule before launch (HBR, A Refresher on A/B Testing).
Quantitative research techniques
The four designs above are about what you can claim. Techniques are about how the numbers get collected. Most studies combine two or three of the data-collection methods below.
- Surveys and questionnaires
- Behavioral and analytics data
- Structured interviews and benchmark tests
Surveys and questionnaires
Surveys quantify opinions, attitudes and self-reported behavior at a scale no other technique reaches. Closed-ended items — rating scales, multiple choice, ranked lists — make responses directly comparable across segments and over time.
Two things decide whether a survey is worth running. The first is question wording: small changes in phrasing, order and answer options move results measurably, which is why Pew publishes its question wording and testing process in full (Pew Research Center, Writing Survey Questions). The second is sampling: a survey answered only by your most engaged users measures your most engaged users, not your market.
In 2026 the most common product setup is an in-product survey triggered from analytics — shown to users who hit a specific event, such as finishing an import or abandoning checkout — so responses arrive with behavioral context attached instead of as a standalone panel blast. Add one open-ended question at the end; it costs almost nothing and gives you the quotes that explain the scores.
Image by Andreas Breitling from Pixabay
Behavioral and analytics data
Behavioral data is the largest quantitative dataset most teams already own and the most underused. Event streams, funnels, cohort analysis and retention curves are quantitative research: they measure real behavior rather than reported behavior, at full population scale, with no recruiting cost.
Cohort analysis is the workhorse. Group users by the week they signed up, or by whether they used a feature, and compare their curves. It underpins descriptive baselines, feeds correlational analysis, and provides the pre/post series a quasi-experiment needs. Its limit is that analytics tells you what happened and never why, so treat every surprising curve as a prompt for a follow-up study rather than a conclusion (NN/g, Analytics and User Experience).
Structured interviews and benchmark tests
Interviews are usually considered qualitative, but a structured interview — identical questions, identical order, coded to a fixed scheme — produces countable data. The same goes for quantitative usability benchmarks: task success rates, time on task, error counts and standardized scores such as SUS, collected the same way each round so they can be compared over time.
The rule is standardization. If the questions change between participants, you have a qualitative study, and you should analyze it as one. If they do not, you can report proportions. NN/g's overview of when to use which method is the cleanest map of where each technique sits on that axis (NN/g, When to Use Which User-Experience Research Methods).
How to choose the right quantitative research method
Start from the sentence you want to be able to write in the report, then pick the design that entitles you to write it.
- "X% of users do Y" → descriptive
- "Users who do X also tend to do Y" → correlational
- "Things changed after we shipped X, and the comparison group did not change" → quasi-experimental
- "X caused Y" → experimental
Then check three practical constraints before you commit:
- Sample size. Can you reach enough people for the effect you care about? If not, a smaller qualitative study will tell you more than an underpowered quantitative one.
- Control. Can you randomly assign? If not, you are running a quasi-experiment, and you should say so.
- Timeline. Longitudinal and experimental designs need calendar time. Descriptive and correlational work usually runs on data you already hold.
The SMART framework is still a useful check on the objective itself: specific, measurable, attainable, relevant and time-bound. And when you are picking software rather than method, the current landscape is covered in our guide to the best AI user research tools. For a definitional refresher, the Interaction Design Foundation's quantitative research topic page is a good, regularly updated reference.
When you need qualitative depth alongside the numbers
Every quantitative finding ends in the same place: you know the size of the problem and not the shape of it. The conversion test won, but you do not know what the losing variant confused. Retention correlates with second-seat invites, but you do not know what the second person unlocks.
That gap is what qualitative follow-up is for, and it is far less work than it used to be. If you already have interview or session recordings, you can paste a transcript into our free AI transcript analyzer and get themes, quotes and citations back without setting anything up — a useful way to put language around a number before you commit to a full study. When the follow-up needs to be a real study rather than a one-off analysis, our pricing page sets out what running moderated AI interviews at volume costs, and the process for turning them into insights is covered in how to synthesize qualitative data.
Sources and further reading
- John W. Creswell & J. David Creswell, Research Design: Qualitative, Quantitative, and Mixed Methods Approaches, SAGE — the standard reference for the four designs and their assumptions.
- Nielsen Norman Group, Quantitative User-Research Methodologies: An Overview.
- Nielsen Norman Group, When to Use Which User-Experience Research Methods.
- Pew Research Center, Writing Survey Questions and its methods hub.
- William M.K. Trochim, Research Methods Knowledge Base: Quasi-Experimental Design.
- Interaction Design Foundation, What is Quantitative Research?.
Conclusion
There are four types of quantitative research, and choosing between them is really a question of how strong a claim you need to make and how much control you have. Describe first, look for relationships second, and reserve experiments for the decisions worth the calendar time. Whichever you pick, write down the claim before you collect the data — and plan the qualitative follow-up that will explain the number once you have it.