PHLT 8032 Week 1 SPSS Setup Walkthrough Example

Setup Walkthrough for a Composite County Adult Health Survey Extract in SPSS: Provenance, Import Repairs, Variable Groups, Special Codes, the Survey Weight and What Was Verified Before Any Analysis

Student Name

Walden University

PHLT 8032: SPSS Revealed

Instructor Name

Month Day, Year

Setup Walkthrough for a Composite County Adult Health Survey Extract in SPSS: Provenance, Import Repairs, Variable Groups, Special Codes, the Survey Weight and What Was Verified Before Any Analysis

Provenance and Unit of Analysis

The file described here is an extract from the Harlow County Adult Health Survey, a composite telephone and web survey of adults aged 18 and older built for this course. The extract covers the 2024 survey year and arrived from the county epidemiologist as a spreadsheet of 1,846 rows and 16 columns. One row is one interviewed adult. Nothing in this walkthrough is an estimate about the county; every number reported is a count used to confirm that the file was set up as intended. Field (2018) treats the Variable View as the place where an analyst decides what each column means before the software is allowed to compute anything, and that order is followed here. A reader should be able to check every decision below against the survey codebook without opening SPSS.

What the Import Got Wrong

The spreadsheet was opened through the Read Text Data route with the first row read as variable names. SPSS assigned types by scanning the opening rows, and two of its guesses were wrong. The respondent identifier, a six-digit code with leading zeros such as 004117, was read as numeric, so 004117 became 4117 and the identifier could no longer be matched to the interviewer log. The interview date was read as a string because a handful of rows used a different separator. The file was imported again with the identifier forced to string width 6 and the date column read as a date in mm/dd/yyyy form, and the seven rows with the odd separator were corrected in the spreadsheet copy before the second import, with the original spreadsheet kept unchanged. Huebner et al. (2016) argue that screening the data systematically before any research question is tested is part of sound research practice, because abnormal values caught late tend to masquerade as findings; a truncated identifier is exactly such a value.

A third problem was less visible. District of residence arrived as typed text, not a code. A frequency run on the string variable showed six distinct entries for a county that has four districts: North, north, Central, River, a River entry with a trailing space, and South. Left as text, any grouping procedure would have produced six groups.

Converting District to a Coded Variable

District was converted with Automatic Recode after the text had been trimmed and case-standardized, producing a numeric variable with value labels 1 North, 2 Central, 3 River and 4 South. The counts before and after are shown in Table 1 so a reader can confirm that no respondent was lost or moved.

The level for district was set to nominal, since the codes name places and imply no order.

Table 1

District of Residence Before and After Conversion

Counts are unweighted rows in the composite extract before any cleaning.

The Variables, Taken in Groups

Setup was done in three groups because the same logic applies within each. The first group holds the identifier, the interview date, district and the survey weight, which describe the record rather than the person. The second holds demographics: age in years, sex, household income bracket and insurance status. The third holds health measures: self-rated general health, two month-long counts of bad health days (one physical, one mental), smoking status, height, weight, weekly minutes of physical activity and influenza vaccination in the past year.

Every column received a short lowercase name and a full label that a reader could understand without the questionnaire, as Pallant (2020) recommends for files that will be shared. Table 2 is an excerpt of the Variable View after setup, limited to the columns a reviewer needs.

Table 2

Variable View After Setup (Excerpt)

The full Variable View holds 16 variables; the measure column is provisional until the definition week.

The Identifier

The identifier is stored as text because it is a label, not a quantity. A mean identifier would be meaningless, and storing it as a string prevents any procedure from treating it as a number. A duplicate check with Identify Duplicate Cases found one identifier, 011392, on two rows with identical answers and interview times one minute apart. It looks like an interview keyed twice, but deleting a row is a cleaning decision, not a setup decision, so both rows were kept and flagged with an indicator variable for the next stage.

Codes That Are Not Quantities

The codebook uses three codes on the two days-of-poor-health items that a careless setup would average as days. Code 88 means none, code 77 means don't know and code 99 means refused. They behave differently. Codes 77 and 99 carry no information about the number of days, so they were declared user-missing in the Missing column, which keeps them visible in frequency tables but removes them from calculations. Code 88 is a real answer: it means zero days. Declaring it missing would discard the largest group of respondents, and leaving it as a value would turn every healthy respondent into someone with 88 bad days in a 30-day month. At setup it was given the value label "None (to recode as 0)" and left in place; the recode itself belongs to the cleaning log, where it can be dated and counted. In this extract, 88 appears 1,032 times on mental health days and 1,104 times on physical health days.

Similar codes on other items were simpler. Self-rated health uses 7 for don't know and 9 for refused, insurance uses the same pair, and income uses 9 for refused; all were declared user-missing because none of them is a real answer to the question asked.

The Survey Weight

The final weight adjusts the respondents so they resemble the county's adult population by age group, sex and district, correcting for groups that answered at lower rates. It was labeled with that purpose and left switched off. Every count in this walkthrough and in next week's definitions is unweighted and describes respondents only. Weight Cases will be turned on in later weeks when a figure is meant to describe county adults, and each such figure will be labeled weighted. Lumley (2010) notes that applying a weight in software that ignores the survey's design changes point estimates but understates their uncertainty, a limit that will be stated whenever weighted figures appear.

What Was Verified, and How

Five checks closed the setup. The identifier was confirmed as unique except for the single flagged pair. A frequency run on district confirmed four groups summing to 1,846. Frequency tables on every coded variable confirmed that labels display in place of numbers, so output reads "Poor" rather than "5." The missing-value settings were confirmed by running Descriptives on mental health days with and without the 77 and 99 declarations; the maximum fell from 99 to 88, showing the declarations work and that 88 still awaits its recode. Finally, the variable list was checked for any string variable holding numbers, and none remained after district was converted.

Two items were carried forward rather than resolved: the duplicate pair and the 88 recode. Both are named here so they cannot disappear between weeks. The syntax pasted from each dialog was saved alongside the data file, giving the cleaning stage a record that starts on the first day.

References

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

Huebner, M., Vach, W., & le Cessie, S. (2016). A systematic approach to initial data analysis is good research practice. The Journal of Thoracic and Cardiovascular Surgery, 151(1), 25-27. https://doi.org/10.1016/j.jtcvs.2015.09.085

Lumley, T. (2010). Complex surveys: A guide to analysis using R. Wiley.

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