Marketing Dissertation Examples

The model for reporting a sample honestly, which sounds dull and is what examiners probe hardest. Chapter 3 (from page 23) records: the survey ran on Amazon Mechanical Turk through a named university lab; participants were quota-limited to a 50/50 gender split and ages 45 to 65 and otherwise randomized; it took 12-15 minutes and paid $5; a pilot of 20 ran in April 2023 and was excluded from the final sample; full collection ran June to early August 2023; 511 responses were collected and 492 analysed, the difference being omitted questions, time outliers and disingenuous ratings.

Every number a reader needs to judge the sample, including the ones that make it look smaller. Most dissertations report the final N and leave the examiner to ask what happened to the rest. It is also a usable model for ratings-based conjoint analysis, with the choice of method justified.

The cleanest hypothesis development section in the collection, and the shortest document to demonstrate it. Section 1.2 (page 9) is subdivided one hypothesis at a time: phygital design and purchase intention, perceived usefulness and playfulness, the mediating role of both, then the moderator.

Each subsection argues from prior theory and ends "Therefore, we hypothesize:" followed by the hypothesis. That is the structure every methods textbook describes and few students see executed: the hypotheses are not a list at the end of the literature review, each one is the conclusion of its own short argument, and a reader can see exactly which literature supports which prediction. It sits on a named theoretical base (the Technology Acceptance Model, plus hedonic and utilitarian shopping motivations) and tests mediation and moderation rather than main effects alone.

Sixty pages, and the best experimental-design example here. Go to section 4.2, "Method" (page 30). Cai needs hotel brands that differ on personality and nothing else.

Real brands carry baggage, so she invents two: both called "Hotel Genova", both priced at $150 a night, differing only in decor and service style, each presented with a description and two photographs. That is controlling a confound by construction rather than by covariate. Then she pretests and reports the checks: the sincere hotel was rated more sincere than exciting (5.67 vs 4.55, p < .001), the exciting one more exciting (6.07 vs 4.73, p < .001), and the two were equally liked overall (5.67 vs 5.82, p = .574), so any later difference cannot be explained by one simply being nicer. A 2x2 between-subjects design follows. Manipulation checks are what student experiments most often skip.

Read section 1.4, "Research design" (page 31), even if you never touch econometrics. It is the clearest explanation in the collection of what an identification strategy is and why anyone needs one.

Cengiz states the problem in one sentence a non-specialist can follow: firms do not advertise randomly, they advertise where and when people were already going to buy more, so a naive analysis credits advertising with sales it did not cause. Then the fix: compare consumers living on opposite sides of a media-market border, who share business cycles, economic conditions and local demand shocks but are exposed to different advertising because firms buy media by market. Then the payoff, stated plainly: the design "leaves only the level of advertising exposure to explain the differences in beer sales between neighboring DMA populations." Problem, design, what the design rules out, and the structure transfers to any field.

The open-science entry, and the one to read if you are running experiments and have been told to preregister. Goenka states that "the hypothesis and study design were preregistered on the open science framework website", in the methods, as a matter of record, before the results. Two other things in the same section are worth copying.

The sample size is justified before collection and on the literature: prior work on binding values typically uses 300-600 participants, so 600 was decided a priori for each study. That is what "a priori sample size" means, and it is a better answer than a power calculation nobody believes. And the procedure includes a filler task, an unrelated word-completion exercise dropped in between the manipulation and the product evaluation, placed there deliberately to reduce demand effects, with that reason stated. Design decisions with their reasons attached, rather than a list of what was done.