Qualitative vs Quantitative Research: How to Choose

Qualitative research collects words, observations and reasons from a small number of people to explain why they think or act as they do. Quantitative research collects numbers from many people to measure how many, how much or how often. The first produces themes and language. The second produces percentages and comparisons.

A quick way to tell them apart is to look at the question you ask. A qualitative question is open: "Walk me through the last time you bought a mobile plan. What nearly stopped you?" A quantitative question is closed: "In the last 30 days, how many times did you recharge? None, 1 to 2, 3 to 5, 6 or more."

There is a second axis that is easy to miss. Attitudinal research captures what people say; behavioural research captures what they do (Nielsen Norman Group). Interviews are qualitative and attitudinal. Surveys are quantitative and attitudinal. Analytics are quantitative and behavioural. Watching someone use a product is behavioural and can be either. Where what people say and what they do disagree, you have usually found the interesting part. A survey alone cannot show you that gap.

Choose by the verb in your question. "Why", "how" and "what does it mean" need qualitative research. "How many", "how much" and "which wins" need quantitative research. If you cannot yet list the answer options, run qualitative first. If you hold a number you cannot explain, run qualitative second.

Nielsen Norman Group describes the same sequence: begin with qualitative work when the topic is unfamiliar or loosely defined, then test or measure what you learned at scale (Nielsen Norman Group on mixed-methods research).

Run three tests before you commission anything.

Qualitative research needs far fewer participants than quantitative research, and the number is set by saturation (the point where extra interviews stop producing new themes), not by a formula. Quantitative research needs a sample sized to the precision you want, which is set by arithmetic.

Qualitative. The best-known test of saturation analysed 60 in-depth interviews with high-risk women, 30 in Accra, Ghana and 30 in Ibadan, Nigeria, on a narrow topic with narrow objectives and a fairly similar group. Saturation was reached within the first twelve interviews. In the Ghana data, 80 of 109 codes (73%) had appeared within the first six transcripts and 100 (92%) by twelve. For high-level themes, six interviews may have been enough (Guest, Bunce and Johnson, Field Methods, 2006). That holds for a uniform group on a focused topic. A mixed audience with several behaviours to compare needs more.

Nielsen Norman Group advises starting with five or six interviews and analysing as you go, with 20 to 30 for diverse populations and as few as five for narrow ones (Nielsen Norman Group). Our recommendation for India: treat each language, city tier or user type you expect to behave differently as its own population, and plan roughly six to twelve interviews for each. Tier 1 and Tier 2 cities are India's grouping of cities by size and income, and behaviour often differs between them.

Quantitative. The margin of error is how far a survey result could sit from the truth, in either direction. The standard formula for a percentage is n = z² × p(1 − p) ÷ e². Here n is the completes needed, z = 1.96 for 95% confidence, p is the expected share (0.5 is the most cautious assumption) and e is the margin of error as a decimal. These are our calculations:

Two cautions apply. The formula assumes a random sample, so for an online panel treat it as a guide to precision, not a guarantee. And to say that two groups differ, you need more than each group's own margin. With 150 completes in each of two groups (our calculation, same assumptions), the gap between them must exceed roughly 11 percentage points before you can trust it.

Quantitative research gets expensive because every segment you want to read separately needs its own sample, and the segments multiply. A cell is one combination you want a result for, such as Tier 2 city by age 25 to 34. Each cell needs its own sample: 150 completes gives about ±8 points, and 385 gives ±5 points.

Here is a worked example, our model from the formula above. Suppose you want to know why 18 to 45 year olds are not switching to your product, and how big the problem is in each city tier.

• Quantitative. 3 city tiers by 4 age bands gives 12 cells. At 150 completes per cell (±8.0 percentage points each), you need 12 × 150 = 1,800 completes. For ±5 points in every cell you would need 12 × 385 = 4,620. One national figure at ±5 points needs only 385, so the 1,800-complete design is about 4.7 times larger.
• Qualitative. Two focus groups in each of the 3 city tiers gives 6 focus groups. At six to ten people each, that is 36 to 60 participants. Groups would mix ages, so the output is themes by tier, not percentages by age band.

The quantitative design tells you how big the problem is in each cell. The qualitative design tells you what the problem is. Choose by the decision you must make, and treat cost as a limit on how many cells you can read.

Each cut to the cell list cuts the cost. Our recommendation: drop any age band you will not treat differently. Each band removed cuts 3 cells, or 450 completes at 150 per cell. This is our model, not an observed result.

The most common mistakes are judging interviews by survey rules, treating a survey with open boxes as qualitative, and asking people to predict their own behaviour. Each leads to a study that cannot answer the question it was commissioned for.

• Judging interviews by survey rules. Ten interviews are not a poll. Their job is to find the reasons and the words people use. If a stakeholder asks for significance, offer a quantitative follow-up for sizing, and keep the interviews for explanation.
• Calling an open comment box qualitative research. A few free-text answers inside a 1,000-person survey are data to skim. Depth needs a conversation where you can ask "why?" a second and third time.
• Asking what people would do. Our recommendation: watch what people do rather than ask what they would do. Usability sessions, a diary study or observation give you behaviour instead of prediction.
• Going big first. A national survey written before any interviews tends to measure your assumptions with great precision.
• Using qualitative work to size a market. If the decision turns on a percentage, you need a sized sample.

We recruit the people for qualitative studies. A buyer shares a research brief with the objective, method, audience, geography, sample, timeline and recruitment criteria. We assess feasibility and send a tailored proposal and quote, because pricing depends on the method, audience, sample and how hard the profile is to recruit. We do not work from a fixed rate card.

Once scope, commercials and timelines are agreed, we recruit, screen and coordinate participants for the activity: in-depth interviews, focus groups, digital diary studies, online bulletin boards, ethnographic studies, shop-alongs, product testing, usability and app testing, and other formats. We recruit across India, in major metros, Tier 1 and Tier 2 cities, and urban, semi-urban and rural areas, with language needs assessed for each project. Participant incentives are processed study by study after the session is completed and validated.

If you are new to us, we can discuss a pilot or introductory first project, depending on your requirements, so you can judge participant quality before committing to a larger study. For method detail, read our guides on how to conduct focus groups in India and how to recruit participants for a study in India.

To share a brief, contact the Think Design Research Services team that operates Qualfacto or email [email protected], and we will tell you what sample your question needs.