PHLT 8032 Week 10 SPSS Syntax Audit Example

Syntax Audit for the Harlow County Survey Analysis: An Annotated Command File in Labeled Blocks and a Memo on the Full Rerun, Three Discrepancies, Their Causes, Their Fixes and the Figures They Changed

Student Name

Walden University

PHLT 8032: SPSS Revealed

Instructor Name

Month Day, Year

Syntax Audit for the Harlow County Survey Analysis: An Annotated Command File in Labeled Blocks and a Memo on the Full Rerun, Three Discrepancies, Their Causes, Their Fixes and the Figures They Changed

Purpose of the Audit

Every figure reported in this course so far was produced in SPSS sessions spread over nine weeks, some through dialogs and some from pasted syntax. This audit asks a simple question: if the raw spreadsheet were imported again and the syntax file run from top to bottom, would every reported figure come back exactly? Sandve et al. (2013) list keeping a record of every step that produces a result as the first rule of reproducible analysis, and Wilson et al. (2017) recommend a single script that rebuilds the analytic file from the raw data. Before this week, the syntax file existed but had never been run as a whole. It has now been reorganized into labeled blocks, commented, and rerun on a clean import.

The Annotated Syntax File

The file opens with a header comment naming the data source, the survey year, the author, the date of the audit and the rule that the file must run from top to bottom without dialogs. It is divided into seven blocks. Each block begins with a comment stating the decision it carries out and the cleaning log entry or week that justifies it. Table 1 shows the block structure with each block's opening comment and its principal commands.

Comments state reasons, not mechanics. The comment opening Block 3, for example, reads: "88 means none in the codebook, so it is recoded to zero into ment_d and phys_d; the source variables are kept unchanged; see log entries 1 and 2." A reader who opens the file cold can tell why each transformation exists and where the decision was recorded, which is the test Pallant (2020) proposes for a well-documented file. Exploratory runs that informed a decision but produced no reported figure were moved to a final section headed "Not reported," so the main blocks contain only commands behind published numbers.

Table 1

Structure of the Audited Syntax File

Note. Comments use the asterisk form and close with a period; each block ends with EXECUTE where a transformation must be applied before the next block reads it.

The Rerun

The raw spreadsheet was imported into a new session, and the syntax ran from Block 1 to Block 7 without intervention. Every figure reported in the Week 4 descriptives, the Week 5 chart, the Week 7 and Week 8 results and both Week 9 tables was then compared with the rerun output, along with the figures in the working draft of the final report: 47 figures in all. Forty-four matched to the last reported decimal. Three did not, and each is described below with its cause, its fix and the figures it changed.

Discrepancy 1: A BMI Recode Run Twice

The draft final report stated that 31.6% of respondents with a valid BMI were overweight or obese. The rerun gave 68.6%. The cause was a recode that collapsed the four BMI categories into three by overwriting the variable in place: underweight and normal were combined as 1, overweight became 2 and obese became 3. In the original session it was run once through the dialog and again when the pasted syntax was executed. A recode in place is not safe to repeat. The second pass treated the new codes as old ones, so overweight respondents, now coded 2, were moved into category 1, and obese respondents, now coded 3, were moved into category 2 under the label "overweight." The saved file therefore held a category that looked plausible and was wrong. The fix replaces the in-place recode with a recode from the scale BMI variable into a new variable, bmi_cat3, which gives the same result however many times it runs. The corrected distribution is 31.4% normal or underweight, 37.0% overweight and 31.6% obese, so the 31.6% in the draft was in fact the obese share alone.

Discrepancy 2: SELECT IF Where FILTER Was Meant

The draft report gave the share of respondents with frequent mental distress, 14 or more poor days, as 12.3%. The rerun gave 10.9%, the figure reported in Week 4. The cause was a subgroup check run for the Week 7 discussion, which limited the analysis to respondents under 65 with SELECT IF. SELECT IF deletes the unselected cases from the active file, unlike FILTER, which only hides them, and the file was saved afterward. The next session opened a working file holding 1,361 respondents instead of 1,845, and the frequent distress share was computed on that reduced file. Field (2018) warns that the delete option in Select Cases is permanent once the file is saved, which is exactly what happened. The fix restricts SELECT IF in the syntax to the single duplicate removal, uses FILTER BY for every subgroup check with FILTER OFF immediately after, and adds a line after Block 5 that prints the case count so any unexpected loss shows at once. The 12.3% also appeared in the Week 7 discussion post as a figure for all respondents; a correction was posted to the thread.

Discrepancy 3: A Percentage From Before the 88 Recode

The draft report gave the share of respondents with frequent physical distress as 15.8%. The rerun gave 6.2%. The cause was an output file from before the cleaning log's first entries, when 88 had been briefly declared user-missing rather than recoded to zero. With 88 treated as missing, the 1,104 respondents with no poor physical health days dropped out of the denominator, and 113 respondents with 14 or more days were divided by 715 instead of 1,819. The figure had been copied from an old output file rather than regenerated. The fix is procedural: every figure in the final report will now be regenerated from the audited syntax, and output files from before the cleaning log are archived in a folder marked superseded.

What the Audit Changes

Three figures in the draft final report change: the BMI category distribution, frequent mental distress from 12.3% to 10.9% and frequent physical distress from 15.8% to 6.2%. No figure in the Week 4, 5, 7, 8 or 9 write-ups changed; the one discussion post figure was corrected in the thread. The syntax file now reproduces all 47 figures from the raw spreadsheet, and the final report will cite it in an appendix.

References

Field, A. (2018). Discovering statistics using IBM SPSS Statistics (5th ed.). SAGE.

Pallant, J. (2020). SPSS survival manual: A step by step guide to data analysis using IBM SPSS (7th ed.). Routledge.

Sandve, G. K., Nekrutenko, A., Taylor, J., & Hovig, E. (2013). Ten simple rules for reproducible computational research. PLoS Computational Biology, 9(10), Article e1003285. https://doi.org/10.1371/journal.pcbi.1003285

Wilson, G., Bryan, J., Cranston, K., Kitzes, J., Nederbragt, L., & Teal, T. K. (2017). Good enough practices in scientific computing. PLoS Computational Biology, 13(6), Article e1005510. https://doi.org/10.1371/journal.pcbi.1005510