Data Analysis Using SPSS: From a Codebook to a Reproducible Workflow

1. Define outputs before opening SPSS

Data processing means more than editing a few Data View cells. The aim is an analysis dataset with clear provenance, a rerunnable command file and results matching the research question.

Before importing data, document dependent, explanatory and control variables, score scales and exclusion rules in the codebook. Subsequent auditing becomes nearly impossible if these decisions exist only in the analyst's memory.

2. Organize folders and protect raw data

Keep raw data read-only. All edits should generate derived data. This follows reproducible-research principles separating data, code and results (Wilson et al., 2017).

Use stable folder and filenames. Avoid names such as final, final2 or final_moi, which do not identify which file produced the thesis results.

3. Turn the questionnaire into a codebook

Variable View is the codebook's technical representation, but maintain an independently readable codebook table. Each row describes a variable. Minimum columns are variable name, question, data type, value labels, missing codes, valid range and analytical role.

Distinguish system-missing from user-missing. System-missing occurs when SPSS has no numeric value. User-missing is an intentional code, such as 99 for not applicable. SPSS handles it correctly only after the code is declared (IBM, n.d.-b).

4. Import data and check structure

After import, check rows, variables, types, duplicate IDs and ranges before calculating scale scores. A file opening successfully does not establish correct import.

The following commands illustrate a workflow using a simulated CSV file. Adapt filenames and coding rules to the actual codebook.

GET DATA
  /TYPE=TXT
  /FILE='data/raw/survey.csv'
  /ENCODING='UTF8'
  /DELCASE=LINE
  /DELIMITERS=","
  /QUALIFIER='"'
  /ARRANGEMENT=DELIMITED
  /FIRSTCASE=2
  /IMPORTCASE=ALL.
DATASET NAME survey.

DISPLAY DICTIONARY.
FREQUENCIES VARIABLES=gender age
  satisfaction1 TO satisfaction4
  /STATISTICS=MINIMUM MAXIMUM.

SORT CASES BY id.
MATCH FILES FILE=* /BY id /FIRST=first_id.
FREQUENCIES VARIABLES=first_id.

5. Declare and assess missing data

Define missing values before scoring. Treating code 99 as valid on a 1-to-5 scale distorts means, standard deviations and correlations.

Missing Value Analysis describes missing patterns and offers diagnostic procedures, subject to SPSS version or module availability (IBM, n.d.-c). Do not infer missingness mechanisms from a single percentage table. Examine collection design, skip logic and associations between missingness and observed variables.

MISSING VALUES gender age
  satisfaction1 TO satisfaction4
  trust1 TO trust4 outcome (99).

FREQUENCIES VARIABLES=gender age
  satisfaction1 TO satisfaction4
  trust1 TO trust4 outcome
  /ORDER=ANALYSIS.

MVA VARIABLES=satisfaction1 TO satisfaction4
  trust1 TO trust4 outcome.

6. Create derived variables through syntax

Use RECODE with INTO to create new variables when original values must be preserved. COMPUTE generates averages, sums or interactions. After each transformation, run FREQUENCIES or DESCRIPTIVES to check ranges.

The example uses MEAN.3, creating a satisfaction score only with at least three of four valid indicators. This illustrative rule is not a universal standard. Determine the minimum from scale structure and the analysis plan.

RECODE satisfaction2 (1=5) (2=4) (3=3) (4=2) (5=1)
  INTO satisfaction2_r.
VARIABLE LABELS satisfaction2_r
  'Satisfaction item 2, reverse scored'.

COMPUTE satisfaction_mean =
  MEAN.3(satisfaction1, satisfaction2_r, satisfaction3, satisfaction4).
COMPUTE trust_mean =
  MEAN.3(trust1, trust2, trust3, trust4).
EXECUTE.

DESCRIPTIVES VARIABLES=satisfaction_mean trust_mean
  /STATISTICS=MEAN STDDEV MIN MAX.

7. Check descriptives and reliability before the main model

Examine distributions, out-of-range values and indicator relationships before modeling. Cronbach's alpha does not replace assessing scale structure; it forms only part of reliability evidence.

Indicator removal requires substantive, measurement-quality and diagnostic reasons. Do not remove indicators solely for a slight alpha increase.

EXAMINE VARIABLES=satisfaction_mean trust_mean outcome
  /PLOT=HISTOGRAM NPPLOT
  /STATISTICS=DESCRIPTIVES
  /MISSING=LISTWISE.

RELIABILITY
  /VARIABLES=satisfaction1 satisfaction2_r satisfaction3 satisfaction4
  /SCALE('Satisfaction') ALL
  /MODEL=ALPHA
  /STATISTICS=DESCRIPTIVE SCALE CORR.

8. Estimate the model while preserving the research question

The regression below illustrates data flow only. Model, standard error and control-variable choices require research-design justification. Report coefficients, confidence intervals and conditions of application, rather than only p-values.

Command and subcommand order follow specific syntax rules. Consult Command Syntax Reference when commands fail or defaults need checking (IBM, n.d.-a).

REGRESSION
  /STATISTICS=COEFF OUTS R ANOVA CI(95)
  /DEPENDENT=outcome
  /METHOD=ENTER satisfaction_mean trust_mean age
  /RESIDUALS=HISTOGRAM(ZRESID) NORMPROB(ZRESID)
  /SAVE=PRED RESID.

9. Rerun from the start and retain the audit trail

Close the dataset, open a fresh SPSS session and run syntax from the first line. If output cannot be recreated, the workflow is incomplete. Retain original output and a PDF for a durable format.

Pasting dialog selections into Syntax Editor is useful for learning commands. Read the syntax before running because dialog defaults may differ from the analysis plan.

10. Reporting template and handover checklist

• Raw data have not been overwritten.
• Codebook matches the questionnaire and Variable View.
• Missing codes are declared before scoring.
• Duplicate IDs, ranges and skip logic are checked.
• All recodes and computations have syntax.
• Case counts are recorded for every analysis.
• Reported tables trace to corresponding output and syntax.
• SPSS version and modules are specified.