| Course | GB892 Doctoral Project III: Data Analysis |
|---|---|
| Unit | Unit 8 |
| Paper type | Inferential results write-up |
| Length | About 870 words, 6 pages double-spaced |
| Format | APA 7 student paper |
| School | Purdue University Global |
| Program | MBA |
| Updated | October 2026 |
Free sample paper for GB892 Unit 8
About Four Percent, Not 3.94: A Unit 8 Inferential Results Write-Up for a Dock Appointment Study
[Student Name]
Purdue University Global
GB892: Doctoral Project III: Data Analysis
Unit 8 Inferential Results Write-Up
[Instructor Name]
[Date]
Haverly Logistics, its centers and every estimate are composite and illustrative; the methods sources cited are real.
Primary Outcome: Dwell
The primary model estimates the change in log dwell at adopting centers after month ten, relative to the change at comparison centers over the same period, with center and month fixed effects. Because only twelve centers supply the comparison, intervals were computed with a wild cluster bootstrap with 9,999 replications, following Cameron et al. (2008).
Dwell at adopting centers fell by about 4 percent relative to comparison centers, with a 95 percent interval of 1 to 7 percent and a bootstrap p value of .012. At the adopters' pre-adoption median of 126 minutes, that is about 5 minutes per visit, with an interval of about 1 to 8 minutes.
The estimate is reported as about 4 percent rather than as the model's printed 3.94 percent. At three centers, gate guards enter arrival times by hand when the scanner fails, which the yard system logs. A sensitivity check that adds random error of up to five minutes to those hand-keyed times, repeated across many draws, shifted the estimate by roughly 0.1 point up or down. Reporting two decimal places would imply a precision the measurement does not support.
Secondary Outcome: Share of Visits Over Two Hours
The share of visits lasting more than two hours, the closest measured proxy for exposure to detention charges, fell by 3.2 percentage points at adopting centers relative to comparison centers, with an interval of 0.2 to 6.2 points and p of .041. With adopting centers averaging about 262 live-unload visits a week each in the post-period, that is roughly 8 fewer long waits per center per week. The interval's lower end is close to zero, so the size of this reduction is uncertain.
Segments of the Visit
The same model was run for each segment of a visit. Minutes from gate-in to door assigned fell by about 11 percent, with an interval of 5 to 18 percent. Minutes from door assigned to unload complete, the negative control outcome, changed by 0.5 percent, with an interval of minus 0.4 to 1.4 percent, which spans zero. Minutes from unload complete to gate-out changed by about 1 percent, with an interval spanning zero.
The pattern is what appointments should produce if they work as intended: the reduction comes from the wait for a door, not from unloading or departure. A negative control that does not move, as Lipsitch et al. (2010) explain, strengthens the case that the primary estimate reflects the intervention rather than some other change at adopting centers.
The minutes saved came almost entirely from the wait for a door, which is the only part of a visit an appointment can change.
Checks
The pre-period event-study coefficients, estimated month by month relative to month nine, were small and showed no trend, consistent with the parallel trends check in Unit 4. Placing a placebo adoption date at month five, using only pre-period data, produced an estimate of about 0.4 percent with an interval spanning zero.
Sensitivity
Three sensitivity models were run. Excluding DC-09, the comparison center that added a shift in month five, gave an estimate of about 5 percent, with an interval of 2 to 8 percent, slightly larger as expected. Excluding DC-03 entirely, the adopting center whose migration month is already dropped from the primary model, gave about 5 percent as well. And the originally planned model in raw minutes, before the outcome was changed to log minutes, gave 6.2 minutes, with an interval of 2.3 to 10.0 minutes; the larger figure reflects the influence of the longest waits on a mean, which the log scale reduces.
Reading the Size of the Effect
Five minutes per visit is modest. Across roughly 262 visits a week at each adopting center, it amounts to about twenty-two hours of truck time a week per center, which matters to carriers and drivers even if no single visit feels transformed. The estimate is also an average: the interviews suggest it is larger where crews were rescheduled around appointments and smaller where floor-loaded freight dominates, though the design cannot confirm those differences.
The Estimates Table
A table lists every estimate reported above, with its point estimate, 95 percent interval, bootstrap p value, number of clusters and number of bootstrap replications. Each row also records the outcome scale, log or minutes, so that no reader compares a percentage with a minute count by mistake.
What the Results Do Not Show
The estimates describe the six adopting centers together. They do not show which center benefited most, a question the Unit 6 seminar showed the design cannot answer. They do not measure detention dollars. And because centers chose whether to adopt, the estimate rests on the parallel trends assumption, which the checks support but cannot prove. Bertrand et al. (2004) caution that difference-in-differences estimates with few groups can look more certain than they are, which is why every interval here comes from the bootstrap.
References
Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-differences estimates? The Quarterly Journal of Economics, 119(1), 249-275. https://doi.org/10.1162/003355304772839588
Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2008). Bootstrap-based improvements for inference with clustered errors. The Review of Economics and Statistics, 90(3), 414-427. https://doi.org/10.1162/rest.90.3.414
Lipsitch, M., Tchetgen Tchetgen, E., & Cohen, T. (2010). Negative controls: A tool for detecting confounding and bias in observational studies. Epidemiology, 21(3), 383-388. https://doi.org/10.1097/EDE.0b013e3181d61eeb
What the GB892 Unit 8 assignment asks for
Unit 8 of GB892 asks you to write up your inferential results. Report the primary estimate with its interval and p value, and convert it to units your readers understand. Decide how many digits the data can support, and explain the decision with evidence about measurement error. Report secondary outcomes and any decomposition of the outcome into parts, and say what the pattern implies. Report checks such as pre-period event-study coefficients and placebo dates. Report sensitivity models, saying which way each moved the estimate and why. Provide a table listing every estimate with its interval, test, cluster count and scale. Discuss the size of the effect in practical terms without overstating it. Close with what the results do not show. Cite the inference methods used. About six pages under headings for primary, secondary, segments, checks and sensitivity is common.
Inside the GB892 Unit 8 example
The write-up opens with the primary model and the wild bootstrap, citing Cameron and colleagues, then reports about 4 percent, interval 1 to 7, near 5 minutes at the 126-minute median. A hand-keying sensitivity check moves the estimate by about 0.1 point, justifying the rounding. The two-hour share falls 3.2 points, translated into about 8 fewer long waits per center per week in a margin note. Segment models show door waits falling about 11 percent while unload time, a negative control, stays flat, and the highlighted sentence explains the pattern with Lipsitch and colleagues. A placebo date shows no effect. Sensitivity models and the raw-minute model follow, then a section on effect size, the estimates table and Bertrand and colleagues' caution.
GB892 Unit 8 rubric: what earns full points
GB892 instructors grade the inferential write-up on accuracy, honesty about precision and completeness. The primary estimate must be reported with its interval and translated into meaningful units. Justifying the number of digits with evidence about measurement error shows unusual care and earns credit. Secondary outcomes and segment results should be reported with their intervals and interpreted cautiously. Checks such as placebo dates and pre-period coefficients are weighted because they test the design. Sensitivity models should be explained by direction and reason. A complete estimates table is expected. Discussing practical size without overstatement shows judgment. Stating what the results do not show, and citing inference methods, completes the work.
GB892 Unit 8 help from the desk
Many GB892 results sections print estimates to more decimal places than the measurement supports. Others report a primary estimate without converting it into units managers understand. Students often report sensitivity models without explaining why each moved the estimate. Negative controls and placebo checks are omitted. Practical size is either exaggerated or ignored. Intervals from conventional errors are reported despite few clusters. Report the estimate with its interval and in practical units, justify your precision, add secondary and segment results, run placebo and pre-period checks, explain each sensitivity model, tabulate everything, discuss size soberly and state what the results cannot show. If your models have run, our team can help you write the results.
Get a GB892 Unit 8 example written to your instructions
Include your model output, the measurement behind your outcome and the Unit 8 prompt and rubric; a note on how the data were recorded helps most. Our sample write-up reports each estimate with the precision that recording supports, intervals first, converted into units a manager would recognize. It is free as a first custom sample and ready in 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 GB892 papers and MBA samples
- GB892 Unit 1: Discussion Board Post
- GB892 Unit 2: Data Preparation Record
- GB892 Unit 3: Codebook
- GB892 Unit 4: Assumption Check Summary
- GB892 Unit 5: Descriptive Results Table
- GB892 Unit 6: Seminar Reflection
- GB892 Unit 7: Thematic Analysis
- GB892 Unit 9: Departure Log
- GB892 Unit 10: Results Chapter Draft
- GB891 Unit 8: Fidelity Monitoring Memo
- GB500 Unit 8: Business Ethics Discussion Post
- GB890 Unit 8: Purpose Statement Exercise
- GB512 Unit 8: Formal Business Report
GB892 Unit 8 questions, answered
How many decimal places should an effect estimate carry?
No more than the measurement can support. Timing recorded to the minute, some of it typed by hand, does not justify hundredths of a percent. A quick check, adding plausible recording error and rerunning the model, shows how much the estimate moves and gives a defensible reason for the rounding chosen. APA style sets formatting conventions, not substantive precision.
Should the write-up report p values at all?
Most sections expect them, alongside intervals and effect sizes rather than instead of them. Leading with the interval keeps attention on magnitude, which is usually the practical question. Report exact values to three decimals where your style guide asks, and avoid treating a threshold as a verdict; results at p of .041 and .052 differ very little.
Why report a wild bootstrap interval instead of the usual one?
With few clusters, twelve here, conventional clustered standard errors tend to be too small, so intervals come out too narrow and tests reject too often. The wild cluster bootstrap performs better in that setting. Reporting the conventional interval in a note lets a reader see the difference, which is usually modest but can matter near a threshold.
Where can I find a free GB892 Unit 8 sample paper?
A complete GB892 Unit 8 inferential results write-up is on this page, reporting a difference-in-differences estimate for a dock appointment portal with bootstrap intervals and checks.
Why report about 4 percent instead of 3.94?
Some arrival times are keyed by hand, and plausible keying error shifts the estimate by about a tenth of a point. Two decimals would claim a precision the data do not support.