MKT 834 Topic 2 causal identification essay example

Data-Driven Marketing Management Grand Canyon University Free custom sample in 24 to 48h

This MKT 834 Topic 2 sample paper asks whether a composite home-improvement retailer's push notifications send customers into stores, putting the question to five designs. It belongs to GCU MKT 834, Data-Driven Marketing Management, listed as MKT-834 and offered in the DBA at Grand Canyon University. A naive comparison shows 19 points, before-and-after and matching rest on unverifiable assumptions, the five-mile boundary supports a checked regression discontinuity, and a randomized holdout finds about 1.3 points. Randomization is the reference standard. The cheapest credible design here, a permanent holdout, is also the most informative one. Each design is judged on what it can rule out.

CourseMKT 834 Data-Driven Marketing Management
TopicTopic 2
Paper typeCausal identification essay
LengthAbout 1,000 words, 2 pages
FormatAPA 7 student paper
SchoolGrand Canyon University (GCU)
ProgramDBA
UpdatedOctober 2026

Free sample paper for MKT 834 Topic 2

1

Did the Notification Send Them, or Were They Already Coming? Five Designs, One Reference Standard and the Assumptions That Earn Causal Standing

[Author Name]

College of Doctoral Studies, Grand Canyon University

MKT-834: Data-Driven Marketing Management

Topic 2 Essay

[Instructor Name]

[Date]

The organization, setting and figures below are a composite written as a model document. No real employer, client, colleague or patient is described.

What this page is doingThe title states the essay's question and position. The home-improvement retailer and its figures are a composite; the sources are real.
2

Did the Notification Send Them, or Were They Already Coming? Five Designs, One Reference Standard and the Assumptions That Earn Causal Standing

The Question and the Position

A composite home-improvement retailer sends push notifications through its app to customers within five miles of a store, announcing weekend promotions. The marketing team believes the notifications send customers into stores. The question is whether they do, or whether they reach customers who were already on their way. This essay puts that question to five designs, from a naive comparison to a randomized holdout, and defends a qualified position: randomization is the reference standard, and other designs earn causal standing only through an assumption that is stated and checked. Figures are illustrative.

The Causal Question

Imbens and Rubin (2015) frame causal questions in terms of potential outcomes: for each customer, there is the visit behavior that would occur if notified and the behavior that would occur if not, and the causal effect is the difference. Only one of the two can be observed for any customer. Every design is a way of estimating the unobserved outcome, and each does so by assuming something about why customers who were and were not notified differ.

Design One: Naive Comparison

Customers who received notifications visited a store that weekend at about 31 percent; customers who did not receive them visited at about 12 percent. The naive estimate is a 19-point effect. Its implicit assumption is that recipients and non-recipients are alike except for the notification. They are not: recipients live within five miles of a store and have the app with notifications enabled, which marks them as frequent, nearby customers who visit often anyway. The assumption fails and cannot be checked away.

Design Two: Before and After

Comparing recipients' visits in weekends before and after notifications began isolates change within the same customers. Its assumption is that nothing else changed between the periods. Notifications began in spring, when home-improvement visits rise seasonally, so the assumption is doubtful; it can be checked by comparing the same seasonal change in a prior year, but the design remains fragile.

Design Three: Matched Comparison

Matching recipients to non-recipients with similar past visit frequency, distance and spending, and comparing their visits, removes differences on those measured characteristics. Its assumption is that no unmeasured difference remains, such as a customer's current project. The assumption cannot be fully checked, because unmeasured differences are, by definition, unmeasured. Gordon et al. (2019) found that even sophisticated matching on rich data often failed to recover effects measured by experiments.

Design Four: The Five-Mile Boundary

The five-mile rule offers a sharper design. Customers just inside five miles receive notifications; customers just outside do not, and they are otherwise nearly identical in distance and, plausibly, in behavior. Comparing visits on either side of the boundary is a regression discontinuity design (Angrist & Pischke, 2009). Its assumption is that nothing else changes at exactly five miles, which can be checked by confirming that past visit rates and customer characteristics are smooth across the boundary. Its limit is that it estimates the effect only for customers near five miles, not for those who live next to a store.

Design Five: Randomized Holdout

Withholding notifications from a random 10 percent of eligible customers for six weeks and comparing their visits with the rest makes the two groups alike in expectation on every characteristic, measured or not. Its assumption is only that the randomization was carried out correctly and that groups did not influence one another, both of which can be checked. In the retailer's holdout, notified customers visited at about 14.2 percent per weekend against about 12.9 percent for the holdout, an effect of about 1.3 points, a small fraction of the naive 19.

Table 1

Five Designs and Their Assumptions, Illustrative

DesignEstimateKey assumptionCan it be checked?
Naive comparisonAbout 19 pointsRecipients and non-recipients alikeNo; plainly false
Before and afterNot reliable hereNothing else changedPartly, against prior years
Matched comparisonDepends on matchingNo unmeasured differencesNo
Five-mile boundaryEffect near five milesNothing else changes at the boundaryYes, by smoothness checks
Randomized holdoutAbout 1.3 pointsRandomization done correctlyYes

What the Retailer Should Do

The retailer should keep a small randomized holdout running permanently, withholding notifications from a random few percent of eligible customers, so that every promotion's effect can be measured against it. It should use the five-mile boundary as a cross-check, since the boundary design costs nothing to run and tests whether the holdout's estimate holds for customers farther from stores. And it should stop reporting the naive comparison, which credits notifications with visits that frequent customers would have made anyway and overstates the program's value many times over.

The Cost of Each Design

The designs also differ in cost. The naive and before-and-after comparisons are free but unreliable. The boundary design is free but limited in scope. The holdout costs the visits the withheld customers would have made because of notifications, which, at about 1.3 points on a tenth of customers, is small. The cheapest credible design is therefore also the most informative one. That is not always true, but when it is, there is little reason to rely on anything weaker.

Why Randomization Is the Reference

Randomization is the reference standard because its assumption is the weakest and the easiest to verify. Other designs can approach it, and the boundary design in particular can be credible, but only by adding an assumption that must be stated and tested. A design whose assumption cannot be checked, such as matching on observed data alone, should be treated as suggestive, not causal.

Conclusion

The notifications appear to send some customers into stores, but far fewer than the naive comparison implies: about 1.3 points per weekend rather than 19. The difference is the gap between customers who were coming anyway and those the notification moved. Randomization establishes that most cleanly; other designs earn causal standing only to the extent that their assumptions are stated and survive checking.

References

Angrist, J. D., & Pischke, J.-S. (2009). Mostly harmless econometrics: An empiricist's companion. Princeton University Press.

Gordon, B. R., Zettelmeyer, F., Bhargava, N., & Chapsky, D. (2019). A comparison of approaches to advertising measurement: Evidence from big field experiments at Facebook. Marketing Science, 38(2), 193-225. https://doi.org/10.1287/mksc.2018.1135

Imbens, G. W., & Rubin, D. B. (2015). Causal inference for statistics, social, and biomedical sciences: An introduction. Cambridge University Press.

Reading the GCU MKT 834 Topic 2 instructions

The Topic 2 assignment in MKT-834 asks doctoral students to evaluate designs for causal claims in marketing. Strong essays frame the question in terms of potential outcomes, apply several designs to the same case, state each design's key assumption and whether it can be checked, and compare the estimates. Students should defend a position on the role of randomization, explain how other designs can earn causal standing, consider costs and recommend how the organization should measure the effect going forward. The essay should also consider what each design costs to run.

How this MKT 834 Topic 2 example is built

Five answers to one question structure the finished essay, each tied to the assumption it needs. Rubin's potential-outcomes framework sets the terms: the effect for any customer is the difference between two outcomes, only one of which can be observed, which Holland called the fundamental problem of causal inference. A naive comparison of notified and unnotified customers assumes they were alike, which app-settings data contradict. Matching on purchase history assumes nothing unrecorded differs, and Gordon and colleagues' comparison of observational methods against large Facebook advertising experiments is cited as evidence that such adjustment often misses the experimental answer. Difference-in-differences around a staggered rollout by region assumes parallel trends. A spending tier at which notifications began supports a discontinuity design near the cutoff. The randomized holdout closes the sequence, with its own limits stated.

Where the marks sit in the MKT 834 Topic 2 rubric

Faculty reward an essay that attaches an assumption to every estimate. An Excellent essay frames the question with Imbens and Rubin's potential outcomes. Credit goes to showing why the naive 19-point gap reflects frequent nearby customers and why matching may fail, with Gordon and colleagues. The rubric values a regression discontinuity at the five-mile boundary, following Angrist and Pischke, with smoothness checks. Strong essays table five designs with their assumptions. Top marks go to a permanent holdout and noting that the cheapest credible design here is also the most informative.

MKT 834 Topic 2 help: mistakes that cost marks

Essays on this topic give away the most when they declare experiments the only valid method and dismiss every alternative, which ignores the designs that economists and marketers use credibly when randomization is unavailable. The mirror error treats a large observational dataset as self-validating, as though volume could stand in for identification. Many drafts name difference-in-differences or discontinuity designs without stating the assumption each needs, so the reader cannot judge whether it holds in the case. Opt-in selection is often missed, even though customers who allow notifications differ from those who do not in exactly the behavior being measured. The Facebook comparison is sometimes cited as proving observational methods always fail, which overstates a finding about particular methods and campaigns. Leaving external validity unanswered lets an examiner reopen the whole position.

Get an MKT-834 Topic 2 example written to your instructions

Send the MKT-834 Topic 2 instructions and the rubric posted in your classroom, with the case or data your section assigned. We write a custom example to those criteria, with the causal question framed as a missing outcome, each design tied to its assumption, the evidence for that assumption examined and the external-validity objection answered, in 24 to 48 hours. The first one is free.

More MKT 834 papers and DBA samples

MKT 834 Topic 2 questions, answered

What is the fundamental problem of causal inference?

Paul Holland's name for the fact that a causal effect for one unit compares two outcomes, with and without a treatment, and only one can ever be observed. Every causal method is therefore a way of estimating the missing outcome from other units. Randomization does this by making treated and untreated groups alike on average. The example uses the idea to judge each design by how well it fills the gap.

When is difference-in-differences credible?

When the treated and comparison groups would have followed parallel paths had the treatment never happened. That cannot be proven directly, but it can be examined by checking whether the groups moved together before the treatment began and whether anything else changed at the same moment. The example applies both checks to the regional rollout and states how its conclusion depends on them.

Did the Facebook experiments show observational methods never work?

No. Gordon and colleagues compared several observational approaches against the results of large randomized advertising experiments and found that the observational estimates frequently differed from the experimental ones, often overstating the effect, even with rich data. That is strong evidence for caution with those methods in advertising, and the example cites it as such rather than as a proof about every non-experimental design.

Where can I find a free MKT 834 Topic 2 sample paper?

The essay is printed here complete: the question, potential outcomes, five designs, what the retailer should do, the cost of each design, why randomization is the reference and the conclusion. Tell us which campaign your seminar examines and a causal essay follows at no cost.

Why is the naive estimate so large in MKT 834 Topic 2?

Notification recipients live near stores and keep app alerts on, so they visit often anyway. The randomized holdout puts the notifications' effect at about 1.3 points, not 19.