BUS 499 Module 5 Gathering the Evidence Example

From 3,412 Records and 52 Phone Numbers to a Usable Data Set: Gathering and Organizing the Evidence on Plan Cancellations

Student Name

Business Administration Program, Aspen University

BUS 499: Senior Capstone

Instructor Name

Month Day, Year

From 3,412 Records and 52 Phone Numbers to a Usable Data Set: Gathering and Organizing the Evidence on Plan Cancellations

The method planned in the previous module called for two kinds of evidence about Prairie Shield's lost plan customers: company records for every quarterly plan customer in the study period and interviews with former customers. This module reports how that evidence was gathered and prepared. It covers extraction, cleaning, quality checks, a description of the sample, interview recruitment, the first stage of coding and the decisions made along the way. No hypotheses are tested here; the goal is a data set that the analysis can rely on.

Extracting the Records

With help from the company's office manager, I exported every residential quarterly plan record with a start or renewal date between January 2023 and June 2025 from the billing system, along with visit records from the service system and the callback log. The exports were joined using the customer account number, which was then replaced with a study number. The combined file contained 3,412 customer records. Names, addresses and phone numbers were never copied into the analysis file.

Cleaning the Data

Cleaning proceeded in three steps, each recorded with counts. First, 37 duplicate records, created when customers changed their payment method and were entered twice, were merged. Second, 21 test accounts created by staff during the software change were removed. Third, 168 customers whose cancellation reason was recorded as moving outside the service area were excluded, because the study concerns customers who chose to leave while they could still be served. These three steps removed 226 records, leaving 3,186 customers for analysis.

Checking Data Quality

Three checks were run. Every record was checked for missing values in the outcome and predictors; only the complaint field had gaps. Complaint records for January through April 2023 were missing for most customers, because callbacks were logged on paper before the new software was installed. As planned, these customers remain in the data, and the analysis will be run twice, with and without them. Second, technician assignments were checked against the routing software's visit history, and 14 visits with no technician recorded were filled in from paper service tickets. Third, cancellation dates were checked against billing to confirm that canceled customers were not billed afterward.

Describing the Sample

Overall, 1,054 customers, or about 33% of the sample, canceled within twelve months of starting or renewing. Testing whether that rate differs across the groups above is the work of the next module.

Recruiting and Conducting Interviews

When customers cancel, Prairie Shield asks whether the company may contact them later. Of customers who canceled during the study period and agreed, I selected 52 to reflect variety in billing method, tenure and complaint history. I reached 31 by telephone after up to three attempts, at different times of day. Of those, 19 agreed and completed interviews, 5 declined and 7 asked to be called later but could not be reached again. Interviews lasted between 9 and 31 minutes, averaging 17.

Kvale and Brinkmann (2009) emphasize that the quality of an interview depends on the interviewer's ability to listen, follow up and avoid steering answers. I followed the guide's open questions first, asked follow-up questions only about what customers raised and introduced the specific topics of billing, technicians and complaints only near the end. With permission, 15 interviews were recorded; for the other 4, I took detailed notes and read them back to the participant to confirm accuracy.

Organizing the Interview Material

Saldana (2021) distinguishes first cycle coding, in which the researcher assigns initial codes to segments of data, from second cycle coding, in which codes are compared, combined and organized into categories and themes. After transcribing the first eight interviews, I used descriptive and in vivo codes, the latter using customers' own words, producing an initial codebook of 22 codes such as different person every time, bill felt random, bugs came back and thought it was a one-time fix. The remaining eleven interviews were coded with this codebook, adding three new codes. Following the convergent design described by Creswell and Creswell (2018), themes will not be finalized until the statistical analysis is also complete, so neither form of evidence shapes the other too early.

What the Records Cannot Tell Us

Preparing the data also made its limits clearer. The billing system records how a customer paid but not why they chose that method, so any later finding about billing must allow for the possibility that more committed customers choose automatic payment. The callback log records that a complaint was made and when it was closed, but not whether the customer was satisfied with the result. Technician records show who performed each visit, not how the customer felt about that person. Finally, the records contain no information about customers' experiences with competitors. These gaps are one reason the interviews matter: they can supply the reasons and feelings that the records leave out.

Decision Log

Keeping the Evidence Secure

Throughout collection, the analysis file stayed on the company's password-protected computer, and the key linking study numbers to account numbers was kept in a separate folder that only the office manager could open. Interview recordings were transferred from my phone to that computer on the day they were made and then deleted from the phone. Transcripts used first names only for the interviewer and a participant number for the customer.

Looking Ahead

The data are now ready for analysis. The records are clean and documented, the sample is described and the interview material is coded at a first level. The next module will test the four hypotheses with logistic regression, develop the interview themes and compare the two in a joint display.

Conclusion

Collecting and preparing evidence took two full weeks, more than planned, mainly because of duplicate records and the paper complaint logs. Reporting each step with counts and reasons makes the data set trustworthy and lets a reader follow every decision. The study now has 3,186 customer records and 19 interviews ready for analysis.

References

Creswell, J. W., & Creswell, J. D. (2018). Research design: Qualitative, quantitative, and mixed methods approaches (5th ed.). SAGE.

Kvale, S., & Brinkmann, S. (2009). InterViews: Learning the craft of qualitative research interviewing (2nd ed.). SAGE.

Saldana, J. (2021). The coding manual for qualitative researchers (4th ed.). SAGE.