Qualitative vs Quantitative Data: Which to Use
Consider a hypothetical service that lets customers book appointments online. Its team can count how many people open the booking form and how many reach the confirmation page. That count shows where people leave. It cannot, by itself, tell the team whether the available times are unsuitable, a question on the form is confusing, or some visitors were only checking prices. A conversation with a customer could reveal one of those reasons. That conversation, however, cannot establish how often the reason occurs among everyone who opened the form.
This is the useful distinction between quantitative and qualitative data. Quantitative data measures or counts a defined feature; qualitative data helps explain meaning, experience, and context. The approaches can address different parts of the same problem, and neither is a substitute for deciding exactly what the team needs to learn. Research-methods guidance from Columbia’s Public Health Workforce Training Center describes quantitative methods as better suited to estimating prevalence and the strength of associations, while qualitative methods are better suited to discovering unanticipated factors and understanding situations.
The practical choice is rarely between a spreadsheet and an interview as objects. It is between claims: “How many people encounter this problem?”, “What happens when they do?”, and “Would changing the form fix it?” Those questions need different observations and different safeguards. Asking them in that order prevents a vivid account from being mistaken for a population estimate, or a precise count from being mistaken for an explanation.
The difference lies in what each answer can establish
Quantitative data records values that can be counted or measured against a defined unit: completed bookings, minutes spent on a page, or the share of respondents selecting a response option. It is useful when the decision depends on size, frequency, change over time, or a comparison between groups. To say that bookings fell, the team needs a consistent definition of a completed booking, a time period, and a denominator. Without those, the number has little meaning.
Qualitative data preserves something the predefined fields may miss: a person’s account of why a question felt intrusive, an observed pause at an insurance field, or the sequence of steps someone takes before leaving. Interviews, open responses, observations, and documents can all contain qualitative material. An account can expose a problem the team never thought to include in its survey. It can also show that two customers who both “abandoned” the form were doing different things. One may have failed to find a suitable time; another may have been comparing providers and never intended to book that day. The ability to follow an unexpected lead is a particular strength of qualitative inquiry, as the Columbia methods comparison explains.
The cost follows from those strengths. Standardized questions make responses easier to compare, but they restrict what a respondent can say to the choices anticipated in advance. Open conversations can surface the unexpected, but their flexibility and usually smaller, deliberately selected samples do not yield a reliable percentage for the whole customer base. A transcript is therefore a strong basis for describing a possible mechanism or experience, while a well-designed count is a stronger basis for estimating its reach. Neither conclusion becomes sound simply because more files were collected.
It is also possible to collect both kinds of data in one encounter. A survey might ask customers to choose an appointment category and then describe what made booking difficult. The category can be tallied; the description needs interpretation. Calling the whole survey “quantitative” because most questions have checkboxes would discard the value of the written answer. Calling it “qualitative” because it contains an open box would obscure the useful counts. The unit of analysis and the question asked of each answer matter more than the format of the form.
A digit in a dataset is not always a measurement
The everyday contrast between words and numbers is a helpful start, but it breaks down when a form assigns numbers to categories. A customer record might label the chosen appointment type “1,” “2,” or “3.” Those digits are identifiers. Adding them or averaging them would have no meaning. Penn State’s guide to variable types separates categorical variables from measurements and distinguishes nominal categories, ordinal categories, discrete counts, and continuous measurements.
Nominal categories name alternatives with no inherent order. Appointment type is one example: a consultation is not “more” of an appointment than a follow-up because one receives a larger database code. A count of customers in each category is meaningful, but a mean category code is not. This is one way a quantitative analysis can involve categories that are sometimes called qualitative variables in statistics. The labels describe the variable’s scale; they do not turn a customer survey into an interview study.
Ordinal categories have an order without a guaranteed distance between steps. Consider a satisfaction item with choices from “very dissatisfied” through “very satisfied.” Responses can be ranked, and the team can report the number choosing each option. Yet the distance between “very dissatisfied” and “somewhat dissatisfied” is not known to equal the distance between “somewhat satisfied” and “very satisfied.” The CDC’s guidance on ordered response options makes that distinction explicit and recommends reporting the number or percentage selecting each option. Treating codes such as 1 through 4 as equally spaced measurements needs a separate justification; printing an average with a decimal place does not supply one.
Discrete quantitative values are counts, such as the number of booking attempts per customer during a defined week. Continuous values measure an amount along a scale, such as elapsed time to finish the form. Even here, the definition matters. A timer starting when a tab opens may measure an idle tab as well as work on the form; that is a measurement problem before it is a statistical one. Penn State’s variable-type explanation treats counts and continuous measurements differently because their possible values and interpretation differ.
These distinctions shape the display that a reader can trust. For appointment types, show category counts or shares. For satisfaction choices, show the distribution across ordered options before compressing it into a single score. For elapsed time, state which events start and stop the clock. If a dashboard blends all three into “numbers,” it hides the choices that determine what those numbers mean.
Write the question before choosing a method
Suppose the booking team says it wants to know “why customers do not finish.” That phrase could mean several different jobs. It might want to describe the size of the drop-off, discover what customers experience at the point of departure, compare completion across devices, or determine whether changing a field causes more people to finish. Those are not interchangeable questions. Research-question guidance connects a focused question to the design and variables used to answer it, and distinguishes descriptive, comparative, and relationship questions from exploratory qualitative ones.
For a descriptive question, such as “What share of people who start the form complete a booking?”, use a count with an explicit population and period. In a purely illustrative calculation, if 240 of 800 form starts became completed bookings during a specified week, the completion share would be 240/800, or 30%. That figure describes those 800 starts under the stated definition. It does not say why the other 560 starts did not finish, nor whether 800 starts represent 800 distinct people. If repeated visits matter, the team must choose whether the unit is a person, a session, or a form start before reporting the result.
For an exploratory question, such as “What do customers think the insurance field is asking, and what do they do when uncertain?”, observation or an open interview has a better chance of exposing the sequence. A customer might describe a concern that the team’s fixed answer list did not anticipate. The purpose is to understand the range and workings of a problem. A few strong descriptions can justify revising a question or investigating a possible obstacle, but they cannot justify a claim that most customers have the same concern.
For an association question, the team might compare completion rates for visitors using different devices. A difference would show that device and completion occur together in the observed data. It would not establish that the device caused the difference. Visitors on different devices may be booking under different circumstances; a mobile visitor may be interrupted more often, or the available times may vary. The data collection and comparison plan must account for plausible alternatives before the team makes a causal claim.
A causal question asks what would happen if the service changed something: for example, whether moving a confusing question later in the form increases completed bookings. The team would need a suitable comparison between the changed and unchanged experiences, with the outcome measured consistently. Interviews could help identify the field to change and explain reactions afterward, but interviewing alone would not measure the effect of the change. Likewise, a before-and-after count could be misleading if availability or visitor mix changed at the same time. The research question sets the design; the words qualitative and quantitative do not do that work on their own.
Collect people and observations that match the claim
Once the question is clear, the next decision is whom and what to observe. To estimate how widespread an obstacle is, the team must define the population it cares about: all visitors, people who started the form, eligible customers, or those who tried to book a particular service. Those populations can give very different answers. A survey of people who already completed booking will miss people who left before seeing the invitation. Thousands of responses cannot repair that coverage gap merely by making the spreadsheet large. Columbia’s comparison of methods notes that generalizing a quantitative result depends especially on how the sample was obtained, including probability-based sampling where appropriate.
To understand the booking experience, deliberate selection can be more useful than a small group chosen to resemble the whole customer base. The team could speak with customers who completed, customers who left at different steps, and people using different devices; those cases may expose different pathways. This is a way to explore meaning and process, not to estimate the share of all customers who struggled. A review of qualitative research methods emphasizes choosing collection methods and participants for the question and making that choice transparent. The interview group should therefore be described by the experiences it covers and the experiences it may have missed.
The collection method also changes what people can reveal. A structured survey gives every respondent a comparable prompt and response set. It is efficient for a defined set of possibilities, but it cannot capture an omitted answer unless the design leaves space for one. A semi-structured interview uses prepared topics while allowing follow-up questions. Observation can reveal a pause, a return to a previous screen, or a workaround that a customer might not mention afterward. A focus group can show how people discuss a shared service in one another’s presence, though the group setting may also shape what someone is willing to say. Documents such as help requests can preserve a problem in the customer’s own words, but they represent the people who chose to contact support. These are reasons to choose among sources, not reasons to treat any source as automatically representative.
Collection needs a stopping rule that matches its purpose. For a quantitative estimate, define the denominator, period, and comparison before reading the result; a stable measurement rule makes later numbers comparable. For qualitative inquiry, an early account may reveal another kind of experience worth examining. A review of qualitative methods treats transparent collection and analysis as part of rigor. The team can adjust who it seeks next while recording why, rather than pretending the sample was fixed by a universal interview count. Flexibility helps only when the final account shows how new observations affected the interpretation.
Analyze the count and the account on their own terms
For quantitative data, the first task is often more basic than choosing a statistical test. Check what was counted. Did a completion event fire once per booking or again when the confirmation page reloaded? Was a form start recorded for everyone or only for visitors who accepted tracking? Were canceled bookings included? In the illustrative 240-of-800 calculation, changing any of those definitions could change the reported 30% without a single customer changing behavior. The denominator should travel with the percentage wherever it is shown.
The next task is to show the distribution relevant to the decision. A single overall rate may conceal the fact that most starts occur on one device or for one appointment type. A breakdown can make the question more precise, but a small subgroup should not be presented with the same confidence as a well-covered population. For ordered survey responses, the CDC’s response-option guidance supports showing how many people chose each category. That lets a reader see whether satisfaction is concentrated in the middle, split across extremes, or genuinely clustered at one end. An average code can conceal those different patterns.
Qualitative analysis starts by keeping an account attached to its setting. A remark about an insurance field means something different from a first-time visitor than from someone who just saw a rejection message. Researchers can compare observations, label passages, and develop themes, but a theme should explain a useful distinction rather than collect similar words. If one person says the field is confusing and another says the available choices are wrong, a single label of “insurance issue” hides two different repairs. The qualitative-methods review describes transparent analysis as a quality condition. Preserve enough of each case that a reader can see how the interpretation was reached.
The researcher is part of that interpretation. An interviewer employed by the service may hear something different from an independent interviewer, especially when a customer worries that criticism could affect an appointment. Record who asked, how people were recruited, which cases were sought, and why an interpretation changed. The qualitative-methods review treats reflection on the researcher’s role as part of rigor. These details do not make an interview count representative; they help a reader judge which experiences the account can explain and where the team’s own assumptions may have shaped it.
Counting coded comments can be useful for organizing a qualitative project, but the resulting fraction does not become a population estimate. If six of ten deliberately selected interviewees mention unclear wording, the defensible statement is that six people in that interview set raised it under those interview conditions. The team still needs a suitable measurement and sample to say how common the issue is among all visitors. Conversely, a survey showing that a response is common does not explain what different respondents meant by it. Each analysis should keep the claim within reach of the observations that produced it.
Combine the methods when one answer leaves a decision open
The booking problem may genuinely require both kinds of data. If the team does not know which obstacle to measure, begin with interviews or observation, use the findings to write specific survey or event questions, and then measure how widely each obstacle appears. If a dashboard already shows an unexpected drop at a particular step, start with that quantitative pattern and choose interviews that can explain what happens there. If the service can collect both streams during the same period, analyze each and compare the conclusions. These are established ways of connecting qualitative and quantitative work, described in research on mixed-method integration.
The important step is the connection. A report with a chart section followed by a page of customer quotations has not necessarily answered a combined question. The team needs to show which comments illuminate which counts, whether the populations and time periods match, and what conclusion follows from both. For the booking form, a useful comparison could place a completion rate by step beside the themes from people observed at that same step. That arrangement may reveal that the prominent drop and the most vivid complaint concern different parts of the form. It may also reveal that a supposedly common problem is only visible among customers with a particular appointment type.
Results can reinforce one another, add different pieces, or conflict. The mixed-methods integration paper describes confirmation, expansion, and discordance as possible relationships between findings. Suppose, illustratively, a survey suggests that few respondents find the insurance field difficult, while interviews repeatedly describe difficulty there. The first response should be to inspect who answered the survey, how the question was worded, and whether interviewees were selected because they had encountered that field. The disagreement could reflect different samples, different meanings of “difficult,” or a real issue concentrated in a smaller group. Selecting whichever result feels more persuasive would waste the information in the mismatch.
Mixed methods also cost more than adding an open question to a survey. Someone must design the link between the samples, interpret the different forms of data, and make disagreement visible. Columbia’s methods comparison notes that standardized quantitative work needs substantial planning up front, while qualitative analysis often requires substantial effort after collection. Use both when the decision depends on both reach and reason. If the only decision is whether the measured completion rate crossed a predefined threshold, a properly designed count may be enough. If the immediate decision is what wording customers misunderstand, careful observation and interviews may be the better first investment.
Let the finding determine the next move, within its limits
For the hypothetical booking service, a sensible sequence begins with a precise count of where customers leave, using a stable definition of starts and completions. That prevents the team from redesigning the most noticeable field when the actual departure point is elsewhere. The team can then observe or interview customers at the relevant step, including those who complete and those who do not, to learn what the step requires of them. If several plausible explanations emerge, the team can measure those explanations with questions customers understand or with events that directly represent the behavior of interest. The sequence moves from location, to interpretation, to reach.
The team should act differently depending on what it finds. If the trouble is a label that customers consistently misunderstand, revising the label may be a low-cost response, followed by a consistent comparison of the changed and unchanged booking experience. If the issue is that no suitable appointments are available, rewriting the form may improve clarity without creating the missing slots. If the interviews reveal several distinct barriers, a single “booking friction” score would hide choices that require different fixes. The cost of the preferred path is slower early progress: the team must define its measures, recruit people who can show the relevant variation, and analyze what they say. The benefit is a decision aimed at the actual obstacle rather than the first number or quote that attracted attention.
The final conclusion should retain its scale. “In this illustration, 240 of 800 recorded form starts ended in a completed booking” is a quantitative description under an illustrative definition. “Some interviewed customers understood the insurance field as asking for information they did not have” is a qualitative account if interviews actually produced it; it is not a measured prevalence. “Changing that field increased completion” would require an appropriate comparison of outcomes after a change. Those are three distinct statements, each with its own route to support.
Qualitative versus quantitative data is therefore a choice about the kind of uncertainty in front of you. Use counts and measurements to learn how much, how often, and how patterns compare. Use accounts and observation to learn what people experience and how a process unfolds. When the decision needs both, connect the two deliberately and investigate any conflict. The strongest answer is the one whose method, sample, and interpretation fit the claim you intend to make.