Week 03 · Marketing Research & Analytics
Davood Wadi, PhD
Qualitative research examines the reasons, meanings, and motives that underlie consumer behavior.
Qualitative research is unstructured and exploratory, and is based on small samples. It provides insight into the reasons for behavior.
Qualitative data consist of words, images, and observations.
Quantitative research is structured, and is based on larger samples, typically chosen to represent a population. It measures how often and to what extent a behavior or attitude occurs.
Quantitative data consist of numbers that can be analyzed statistically.
Qualitative findings cannot be generalized to a population, because the samples are small and are not selected to be representative.
The two approaches are complementary. Qualitative research frequently precedes quantitative research, and is also used afterward to interpret quantitative findings.
Sort each study by the approach it takes.
Tap a card, then tap the group it belongs to.
Qualitative research is the principal method of the exploratory designs introduced in Week 2.
It is used to define a problem more precisely and to develop hypotheses for later testing.
It is used to identify the attributes, language, and categories that consumers use, so that later questionnaires reflect them.
It is used to examine topics that consumers are unable or unwilling to report accurately in a structured questionnaire, such as emotional, habitual, or socially sensitive behavior.
Qualitative methods are classified as direct or indirect.
In a direct approach, the purpose of the research is disclosed to participants, or is evident from the questions asked. Focus groups and depth interviews are direct approaches.
In an indirect approach, the purpose of the research is disguised. Projective techniques are indirect approaches.
Indirect approaches are used when participants are unlikely to express their true motives in response to direct questions.
Focus groups, depth interviews, and projective techniques obtain data by questioning participants.
They differ in the number of participants, the degree of structure, and the extent to which the purpose of the research is disclosed.
A focus group is a discussion among a small group of participants, typically six to ten, led by a trained moderator.
A session typically lasts between one and two hours and follows a discussion guide that lists the topics to be covered.
Participants are usually selected to be similar in relevant characteristics, such as product usage or life stage, so that differences in background do not inhibit discussion.
The interaction among participants can produce ideas and reactions that would not emerge in individual interviews.
The group setting also has limitations: some participants may dominate the discussion, others may conform to the majority view, and the moderator's behavior may influence the responses.
Focus groups are increasingly conducted online, which reduces cost and allows participants in different locations to take part.
A depth interview is an unstructured or semi-structured interview conducted with a single participant, typically lasting between thirty minutes and more than an hour.
The interviewer asks follow-up questions, known as probes, to uncover underlying motivations, beliefs, and feelings.
Depth interviews are appropriate for sensitive topics, for complex decisions, and for participants such as senior professionals, who are difficult to assemble in a group.
Laddering is a depth interviewing technique that moves from a product attribute to its consequences and finally to the personal value it serves.
For example, a participant may state that a running shoe is lightweight, that a lightweight shoe allows faster running, that faster running produces a sense of achievement, and that achievement supports self-esteem.
Laddering reveals the connections between product attributes and personal values, which can inform positioning and advertising.
A participant in a depth interview discusses a reusable water bottle. Put the answers in the order that laddering moves through them, from the product attribute to the personal value.
Tap a step, then the number where it belongs.
Projective techniques present participants with ambiguous stimuli and ask them to interpret them.
Participants are expected to project their own motives, beliefs, and feelings into their interpretation.
In association techniques, participants respond to a word or an image with the first word or idea that occurs to them.
In completion techniques, participants complete an incomplete sentence or story, such as “People who shop at discount stores are …”
In construction techniques, participants construct a response to a picture or a cartoon, such as the words spoken by a person in a drawing.
In expressive techniques, participants describe the feelings or behavior of another person, a method known as the third-person technique, or act out a role.
The interpretation of projective responses requires trained analysts, and different analysts may interpret the same response differently.
Each case describes a task given to a participant. Name the projective technique it uses.
Responding to a word with the first word or idea that occurs: an association technique.
Completing an incomplete sentence is a completion technique.
Constructing the words spoken by a person in a drawing is a construction technique.
Describing the feelings of another person is the third-person technique, an expressive technique.
Completing an incomplete story is a completion technique.
Acting out a role is an expressive technique.
Which concept does this case show?
Observation-based methods record behavior rather than asking participants to report it.
They are used when participants cannot accurately recall or describe their own behavior.
Observation is the systematic recording of the behavior of people, objects, and events.
Observation can be structured, with a predefined list of behaviors to record, or unstructured, with the observer recording all relevant aspects of the situation.
Observation can be disguised, when participants are unaware that they are being observed, or undisguised.
Observation can be conducted in a natural setting, such as a store, or in a contrived setting, such as a test kitchen.
Observation can be personal, conducted by a human observer, or mechanical, conducted with devices such as eye-tracking equipment or in-store cameras.
Observation records what people do, but not why they do it.
Ethnography is the study of people in their natural environment over an extended period, through observation combined with interviews.
The researcher seeks to understand behavior from the perspective of the people being studied.
In marketing research, ethnography takes the form of in-home visits, accompanied shopping trips, and diaries that participants complete on their mobile phones.
Ethnography can reveal needs that consumers do not articulate, such as the workarounds they use when a product does not meet their requirements.
It requires considerable time and expertise, and its findings depend on the interpretation of the researcher.
Netnography is ethnography adapted to the study of online communities.
The researcher examines the interactions of consumers in forums, social media groups, and review platforms.
Netnography is unobtrusive, since the conversations occur without the involvement of the researcher.
It is limited to consumers who participate in online communities, who may differ from the broader population of customers.
It raises ethical questions concerning the distinction between public and private spaces, the need for consent, and the quotation of posts that can identify their authors.
Focus groups are appropriate when interaction among participants is expected to generate ideas, as in the early evaluation of product concepts.
Depth interviews are appropriate for sensitive topics, complex decisions, and participants who are difficult to assemble in a group.
Projective techniques are appropriate when participants are unlikely to state their true motives directly.
Observation is appropriate when the behavior itself is the object of study and participants cannot accurately report it.
Ethnography is appropriate when the research requires an understanding of behavior in its everyday context.
Netnography is appropriate when the relevant consumers discuss the topic in online communities.
Each case describes a research problem. Name the qualitative method that is most appropriate.
Interaction among participants is expected to generate ideas in the early evaluation of concepts: a focus group.
A sensitive topic and a complex decision: depth interviews.
Participants are unlikely to state their true motives directly, so an indirect, projective technique is appropriate.
The behavior itself is the object of study, and shoppers cannot accurately report it: observation.
The research requires an understanding of behavior in its everyday context: ethnography.
The relevant consumers discuss the topic in online communities: netnography.
Which concept does this case show?
Discussion: A manufacturer of home cleaning products seeks to understand why consumers who state a preference for environmentally friendly products continue to purchase conventional products. Which qualitative methods are appropriate for this problem, and what are the limitations of each?
Qualitative research produces large volumes of unstructured data, such as interview transcripts, field notes, images, and online posts.
Analysis reduces these data to a set of themes that address the research questions.
The analysis must be systematic, so that others can follow the path from the data to the conclusions.
Coding is the assignment of labels, known as codes, to segments of qualitative data that express a particular idea.
In deductive coding, the codes are defined in advance, often on the basis of theory or previous research, and are listed in a codebook.
In inductive coding, the codes are developed from the data during analysis.
A codebook defines each code and provides examples of segments to which the code applies and segments to which it does not.
Many projects combine the two approaches, beginning with a codebook and adding codes as new ideas emerge from the data.
Sort each coding decision by the approach it follows.
Tap a card, then tap the group it belongs to.
Related codes are grouped into categories, and categories are developed into themes.
A theme is a pattern of meaning across the data that addresses a research question.
For example, codes such as “compares prices online,” “waits for promotions,” and “buys store brands” may be grouped into a theme of price vigilance.
Themes are supported by representative quotations from participants.
The analyst also reports cases that do not fit the themes, since they may indicate limits to the findings.
Qualitative samples are not determined by statistical calculation.
Data collection continues until saturation is reached.
Saturation is the point at which additional interviews or observations no longer produce new codes or themes.
The number of interviews required for saturation depends on the diversity of the participants and the breadth of the research question.
The researcher reports how saturation was assessed, for example by recording the number of new codes that each additional interview produced.
A researcher records the new codes that each interview produces. Interviews 1 to 8 produced 9, 6, 4, 3, 2, 1, 1 and 0 new codes, and interviews 9 to 11 produced none. The research question is narrow and the participants are similar. What should the researcher do?
Qualitative findings are evaluated for trustworthiness rather than for statistical significance.
Intercoder reliability is the extent to which two or more coders, working independently, assign the same codes to the same data.
Triangulation is the use of several data sources, methods, or analysts to examine the same question.
Member checking is the practice of presenting preliminary findings to participants to confirm that they reflect their views.
An audit trail documents the decisions taken during the analysis, so that others can evaluate them.
AI tools are now used both to collect and to analyze qualitative data.
They allow qualitative research to be conducted at a larger scale and at lower cost.
The standards of trustworthiness introduced in this week apply to AI-assisted research in the same way as to research conducted entirely by people.
In an AI-moderated interview, a conversational AI system asks the questions, generates follow-up probes in response to the participant's answers, and records the conversation.
Such interviews can be conducted with many participants simultaneously, at any time of day, and in several languages.
Some participants may disclose more to an AI interviewer on sensitive topics, while others may disclose less.
An AI interviewer cannot observe nonverbal cues in text-based interviews, may fail to pursue an unexpected but relevant answer, and may probe in ways that lead the participant.
Participants are informed that the interviewer is an AI system, and their consent covers the processing of their responses by that system.
Large language models can be used to apply a codebook to transcripts, to propose inductive codes, and to summarize themes across many interviews.
They reduce the time required to code large volumes of text considerably.
They can also assign codes that the text does not support, overlook views expressed by only a few participants, and produce different results when the instructions are worded differently.
The quotations that an AI tool presents as evidence for a theme are verified against the original transcripts, since they may be altered or fabricated.
The researcher first develops a codebook with clear definitions and examples.
Two or more human coders then code a sample of the data independently, using the codebook.
The AI tool then codes the same sample, using the same codebook.
The researcher then measures the agreement between the AI coding and the human coding, and compares it with the agreement between the human coders.
The researcher then examines the segments on which the AI and the human coders disagree, and revises the codebook or the instructions to the AI tool.
The AI tool is used for the full dataset only when its agreement with the human coders is comparable to the agreement among the human coders themselves, and the level of agreement is reported.
A research team plans to code 300 interview transcripts with a large language model. Put the validation steps in order.
Tap a step, then the number where it belongs.
Some language models generate text that describes their reasoning before they produce a final answer. This text is called a reasoning trace.
In human decision research, a think-aloud protocol records what participants say while they decide, and researchers code these verbal protocols.
Reasoning traces can be coded in the same way, with a codebook that defines each category, such as the objective the agent states or the skepticism it expresses toward a sponsored listing.
In a study of AI shopping agents, traces that stated the lowest total price, rather than the lowest price per unit, as the objective preceded less complete information acquisition and less frequent choice of the better-value product (Wadi & Ma, 2026b).
In a study of sponsored listings, skepticism in the traces was coded by an AI coder that had been validated against a human coder, as introduced in this week (Wadi & Ma, 2026c).
Each excerpt comes from the reasoning trace of an AI shopping agent. Code it with the category from the codebook.
The agent states that it will minimize the amount paid, regardless of quantity.
The agent adopts price per unit of quantity as its criterion.
The agent expresses doubt about a listing because its placement may have been paid for.
The agent compares the prices alone, without the quantity in each package.
Price per load is a price per unit of quantity.
The agent questions the listing because the sponsorship label indicates a paid placement.
Which concept does this case show?
A reasoning trace is text generated by the model. It is not a record of the computation that produced the model's choice.
Like a think-aloud report, a trace may omit considerations that influenced the choice or offer reasons after the fact.
Not all providers expose complete reasoning traces, which limits the set of models that can be studied in this way.
Coded traces therefore provide evidence consistent with an explanation, while conclusions about behavior rest on the agent's observed choices.
Discussion: A research agency proposes to replace twenty depth interviews conducted by experienced interviewers with two hundred AI-moderated interviews, coded by a large language model. What would the client gain and lose in terms of depth, scale, and trustworthiness, and which validation procedures should the agency be required to report?
Qualitative research examines the reasons for behavior through small, nonrepresentative samples, and complements quantitative research.
Focus groups and depth interviews are direct approaches; projective techniques are indirect approaches used when participants are unlikely to state their true motives.
Observation, ethnography, and netnography record behavior in its context rather than relying on participants' reports.
Qualitative analysis proceeds from codes to themes, and data collection continues until saturation is reached.
Qualitative findings are evaluated for trustworthiness, through intercoder reliability, triangulation, member checking, and an audit trail.
AI tools allow qualitative research to be conducted at scale, but AI coding must be validated against human coders before it is relied upon.
The reasoning traces of AI agents can be coded as verbal protocols, but they describe the model's stated reasoning, not the computation that produced its choice.