| Course | GB794 Advanced Topics in Marketing |
|---|---|
| Unit | Unit 3 |
| Paper type | Experiment design memo |
| Length | About 890 words, 6 pages double-spaced |
| Format | APA 7 student paper |
| School | Purdue University Global |
| Program | MBA |
| Updated | October 2026 |
Free sample paper for GB794 Unit 3
Forty-Four Stores, Twenty-Two Pairs: A Unit 3 Experiment Design Memo for a Nutrition-Bar Assortment Test
[Student Name]
Purdue University Global
GB794: Advanced Topics in Marketing
Unit 3 Experiment Design Memo
[Instructor Name]
[Date]
Thornbury Market, its stores and every figure, including the simulated power values, are composite and illustrative; the methods sources cited are real.
Memo
To: Vice President of Merchandising, Thornbury Market From: Director of Analytics Re: A twelve-week test of the reduced nutrition-bar set before any chain-wide change
The Decision and When the Test Can Answer It
The decision is whether to cut the nutrition-bar set from 140 items to 90 in every store. The test described here can tell you, by the second week of March, whether the cut changes bar sales, total basket value or the share of loyalty baskets containing any bar, and how confident we can be in that answer. If you would rather decide sooner, the memo explains what a shorter test would cost in certainty.
The Unit 2 review explains why a test is worth running. The best published evidence on assortment cuts comes from one online grocer: category sales rose after its cuts (Boatwright & Nunes, 2001), while a store-level analysis of the same reduction found that households' total spending fell (Borle et al., 2005). Category sales alone could therefore mislead us.
Assignment
All 44 stores take part. Stores will be ranked by last year's bar sales and paired with their nearest neighbor in that ranking, giving 22 pairs of stores with similar bar sales. Within each pair, one store will be assigned to carry the reduced 90-item set and the other will keep the full set. The assignment will be made by a short program with a fixed random seed, recorded in the appendix before launch, so that no one at Thornbury chooses which stores change.
Pairing matters because our stores differ widely in bar sales, from small neighborhood stores to large suburban ones. Comparing stores within pairs removes much of that variation from the comparison and makes a given test length more informative. Imai et al. (2009) showed that pair matching of this kind substantially improves the precision of experiments in which whole sites, rather than individuals, are assigned.
Outcomes
The primary outcome is nutrition-bar dollar sales per store-week. Two secondary outcomes guard against the store-level effect the published research warns about: total basket value for baskets containing any bar or any item from adjacent snack categories, and the share of loyalty-card baskets containing any bar. Shelf space for the category will be held constant in both arms, with freed space filled by facings of the remaining items, so that the test measures the effect of fewer items rather than of less shelf.
Power
With only 22 pairs, the test can detect large changes more reliably than small ones. The appendix script simulated the experiment many times using last year's weekly sales, assuming a twelve-week test. If the reduced set changes bar sales by 3 percent, the test would detect it 41 percent of the time; by 4 percent, 63 percent of the time; by 5 percent, 81 percent; and by 6 percent, 92 percent. A change smaller than about 5 percent could easily go undetected.
With 22 pairs and twelve weeks, the test can reliably catch a change of 5 percent or more and will often miss anything smaller.
That is acceptable for this decision. The merchandising team's own estimate of the benefit of the cut, through lower inventory and fewer out-of-stocks, is equivalent to about a 4 percent sales gain, so a test that reliably detects changes of 5 percent will catch any loss large enough to outweigh the benefit. A sixteen-week test would raise power at 4 percent to about 75 percent, if you prefer more certainty and can wait a month longer.
Analysis Fixed Before Launch
The analysis will compare the reduced-set store with the full-set store in each pair, week by week, and estimate the average difference across pairs with a confidence interval. The same comparison will be made for the two secondary outcomes. No other outcomes will be used to judge the test, and the analysis plan will be filed before launch so that we do not search afterward for a measure that happens to favor one answer (Gerber & Green, 2012).
Risks and the Stopping Rule
Three risks are worth naming. First, suppliers whose bars are removed in test stores may object; slotting agreements will be checked, and suppliers informed that the change is a time-limited test. Second, regular customers of removed bars may complain or shop elsewhere; store managers will log complaints, and the loyalty data will show whether those customers' visits fall. Third, the test could reveal a clear loss early. The stopping rule is that if, after six weeks, total basket value in reduced-set stores is lower by more than 3 percent with a confidence interval excluding zero, the test will end and full sets will be restored. That rule is written now so that it cannot be adjusted later.
What I Need From You
Approval of the pairing and the twelve-week window, a decision on whether to extend to sixteen weeks, and agreement that store managers will not be told the purpose of the test beyond the reduced planogram.
References
Boatwright, P., & Nunes, J. C. (2001). Reducing assortment: An attribute-based approach. Journal of Marketing, 65(3), 50-63. https://doi.org/10.1509/jmkg.65.3.50.18330
Borle, S., Boatwright, P., Kadane, J. B., Nunes, J. C., & Shmueli, G. (2005). The effect of product assortment changes on customer retention. Marketing Science, 24(4), 616-622. https://doi.org/10.1287/mksc.1050.0121
Gerber, A. S., & Green, D. P. (2012). Field experiments: Design, analysis, and interpretation. W. W. Norton.
Imai, K., King, G., & Nall, C. (2009). The essential role of pair matching in cluster-randomized experiments, with application to the Mexican universal health insurance evaluation. Statistical Science, 24(1), 29-53. https://doi.org/10.1214/08-STS274
What the GB794 Unit 3 assignment asks for
The third GB794 assignment asks you to write a memo proposing a field experiment to the manager who will act on it. State the decision and the date by which the test can answer it. Explain why a test is warranted, drawing on the published evidence. Describe how units will be assigned, and justify any matching or blocking you use. Name the primary and secondary outcomes and say why each matters. Hold constant anything that would confound the comparison, such as shelf space. Report power, preferably from a simulation using your own historical data, and say plainly what size of effect the test could miss. Fix the analysis before launch. Name the main risks and write a stopping rule. End with what you need the manager to approve. Five pages to one decision maker, with the power script in an appendix, is a typical format.
Inside the GB794 Unit 3 example
The memo opens with the cut from 140 items to 90 and a March answer date. It cites the category gain and store-level loss at one online grocer to explain why a test is worth running. The assignment section pairs 44 stores by bar sales and uses Imai and colleagues to explain the gain in precision. Outcomes, a constant shelf space rule and the power simulation follow: 41, 63, 81 and 92 percent at 3 to 6 percent changes, with the highlighted sentence stating what the test can and cannot catch. The memo weighs this against the team's expected 4 percent benefit and offers sixteen weeks as an option. Gerber and Green support filing the analysis in advance, and three risks and a six-week stopping rule close it.
Where the points sit in the GB794 Unit 3 rubric
Points in the GB794 design memo depend on whether a manager could approve and run it. Instructors give credit for a decision and answer date stated on the first page. Assignment earns heavy weight when it is random, documented and improved by matching suited to the setting. Outcomes should include a primary measure and guards against the losses the literature warns about. Holding confounds constant, such as shelf space, shows careful design. A power analysis from the organization's own data, with an honest statement of what the test could miss, is often the most important section. Fixing the analysis in advance and writing a stopping rule show integrity. Risks should be specific to the organization, and the memo should end with a clear request.
Common GB794 Unit 3 mistakes, and how to avoid them
Many GB794 experiment memos propose a test without saying how units will be assigned or who decides. Others choose a single outcome that cannot detect the harm the literature describes. Power is frequently missing or calculated with assumed values rather than the organization's own data. Students often promise to detect any effect, which no design can. Analysis plans are left for later, inviting a search for favorable results. Stopping rules are rare. State the decision and date, pair or block units, assign them by code, choose outcomes that include a guard, hold confounds constant, simulate power from history, fix the analysis and write the stopping rule. If you have a test to propose, we can help you draft the memo for your manager.
Get a GB794 Unit 3 example written to your instructions
Describe the decision your experiment should inform, the units you could realistically randomize and any historical data you hold, together with the Unit 3 instructions and rubric. The memo returned sizes the test by simulation, defines treatment as it would really run and sets a stopping rule before launch. No charge for a first sample; allow 24-48h. The paper above is an original model document written by our desk, not a submitted student paper and not an official Purdue University Global document.
More GB794 papers and MBA samples
- GB794 Unit 1: Discussion Board Post
- GB794 Unit 2: Consumer Study Critique
- GB794 Unit 4: Attribution Analysis
- GB794 Unit 5: Measurement Validity Review
- GB794 Unit 6: Seminar Reflection
- GB794 Unit 7: External Validity Memo
- GB794 Unit 8: Field Study Proposal
- GB794 Unit 9: Evidence Summary
- GB794 Unit 10: Marketing Research Paper
- GB793 Unit 3: Construct Critique
- GB792 Unit 3: Event Study Appraisal
- GB791 Unit 3: Standard Setting Analysis
- GB740 Unit 3: Entry and Contracting Memo
GB794 Unit 3 questions, answered
What does power by script mean in a design memo?
Instead of plugging assumed values into a formula, a short program simulates the planned experiment many times using real historical data, adds a chosen effect and records how often the analysis detects it. It suits designs like store pairs, where clustering and correlation over time make textbook formulas unreliable. Instructors usually want the code attached and its assumptions stated in the body.
Is twenty-two pairs enough for a doctoral design?
It is enough if the memo is honest about what it can detect. Here the simulation shows about 80 percent power for a 5 percent change and far less for 3 percent. A design that states its detectable effect, and ties that figure to a business threshold, is stronger than one claiming a large sample it could never assemble.
Does the memo need ethics review language?
A shelf test analyzed through aggregate store sales usually raises few human-subjects questions, but sections often ask the memo to say so explicitly and to note what review the university requires before any data enter a dissertation. Loyalty-card data are a different matter, and the memo flags them as needing a data agreement and whatever review the program specifies.
Where can I find a free GB794 Unit 3 sample paper?
This page carries a full GB794 Unit 3 experiment design memo for a 44-store nutrition-bar assortment test, with matched pairs, a simulated power table and a stopping rule.
Why pair stores before randomizing?
Stores differ widely in sales. Pairing similar stores and randomizing within each pair removes much of that variation from the comparison, so a test of a given length becomes more precise.