A step-by-step guide to preparing multi-item survey scales for analysis
Most business and management surveys measure concepts such as job satisfaction, trust or brand loyalty with several items rather than a single question. Before you can test your hypotheses, these items usually need to be recoded and combined into one scale score. Mistakes at this stage, such as forgetting to reverse a negatively worded item, can quietly lower reliability and distort every result built on the scale. Examiners often ask how scale scores were created, so you need a clear and repeatable process. This post explains when to recode, how to use Recode into Different Variables, how to reverse-code items, how to compute scale scores with Compute Variable, how to report the process, and which mistakes to avoid.
What recoding and computing mean:
- Recoding changes the values of an existing variable. Common uses are reverse-coding negatively worded items, grouping categories (for example, combining small education groups) and creating dummy variables for regression.
- Computing creates a new variable from a formula, such as the mean of several items that measure the same construct.
- In most survey studies, you recode first and compute second. A scale score is only correct if every item points in the same direction.
- For a categorical variable with several groups used as a predictor in regression, you recode it into dummy variables coded 0 and 1. With k groups you need k − 1 dummies, and the group left out becomes the reference category.
Before you start:
- Save a copy of the raw data file and work on a separate file.
- Check the original scale instructions or your codebook to find out which items are negatively worded, and whether scores should be summed or averaged.
- Define missing-value codes, such as 99, in the Missing column of Variable View, so SPSS does not treat them as real answers.
- Run Frequencies (Analyze > Descriptive Statistics > Frequencies) on every item to confirm that all values fall within the response range.
Using Recode into Different Variables:
Use Transform > Recode into Different Variables rather than Recode into Same Variables. The first creates a new variable and keeps your original data. The second overwrites it.
- Move the item into the box, type a new name and label under Output Variable, and click Change.
- Click Old and New Values. Enter each old value and its new value, and click Add after each pair. Then click Continue.
- To group a continuous variable, such as age, into bands, use the Range options in Old and New Values, for example "Range, LOWEST through value". Keep the original variable as well, because grouping loses information.
- Click Paste instead of OK. This writes the command into a syntax file, which you can run, save and repeat later. It also gives you an exact record for your methodology chapter.
Reverse-coding negatively worded items:
Some scales include negatively worded items, such as "I often think about leaving this organisation" in a commitment scale. These must be reversed so that a high score means the same thing on every item.
- On a 1–5 scale, the new values are 1→5, 2→4, 3→3, 4→2 and 5→1. In general, the reversed score equals the scale minimum plus the scale maximum, minus the original score.
- You can enter these pairs in Old and New Values, or use Transform > Compute Variable with an expression such as 6 - Q5 for a 1–5 scale, or 8 - Q5 for a 1–7 scale.
- Give reversed items a clear name, such as Q5_R, and use only the reversed version in the scale score.
- Check the result with a crosstab (Analyze > Descriptive Statistics > Crosstabs) of the original and reversed variables. Every 1 should now be a 5, and so on.
Computing scale scores:
Use Transform > Compute Variable. Type a name in Target Variable and enter the formula in Numeric Expression.
- Mean score: MEAN(Q1, Q2, Q3, Q4, Q5_R) gives the average of the items. It keeps the original 1–5 metric, which makes the results easy to interpret.
- Sum score: SUM(Q1, Q2, Q3, Q4, Q5_R) adds the items. Use this if the original scale requires a total score. Note that SUM() adds only the items that were answered, so a respondent with a missing item gets a lower total.
- Missing items: MEAN() uses whatever items are available, even if only one was answered. MEAN.4(Q1, Q2, Q3, Q4, Q5_R) computes the score only when at least four of the five items are valid. Writing (Q1 + Q2 + Q3 + Q4 + Q5_R) / 5 returns a missing score if any item is missing.
- Reliability first: Before computing, run Analyze > Scale > Reliability Analysis on the items. A negative corrected item–total correlation often means an item has not been reversed.
Worked example: an employee engagement study:
Imagine you survey employees in a services firm. Work engagement is measured with six items on a 1–7 scale, from strongly disagree to strongly agree. Items 3 and 6 are negatively worded, for example "I feel bored at work". The scale instructions say to average the items.
- Step 1: Frequencies show that some respondents have the code 99 for "no answer". You define 99 as missing in Variable View.
- Step 2: You reverse items 3 and 6 with Recode into Different Variables (1→7, 2→6, 3→5, 4→4, 5→3, 6→2, 7→1), creating ENG3_R and ENG6_R. A crosstab confirms that every value was reversed correctly.
- Step 3: You run Reliability Analysis on ENG1, ENG2, ENG3_R, ENG4, ENG5 and ENG6_R, and check that all corrected item–total correlations are positive.
- Step 4: You compute the score: ENGAGE = MEAN.5(ENG1, ENG2, ENG3_R, ENG4, ENG5, ENG6_R). Respondents who answered fewer than five items get a missing score. Finally, Descriptives confirm that the minimum and maximum of ENGAGE lie between 1 and 7.
How to report this in your thesis:
Explain which items were reversed, how scores were computed, how missing items were handled, and how reliable each scale was. If you changed a published scale, for example by dropping an item, say so and give the reason. You can list all items and their coding in an appendix. A sample sentence:
"Items [item numbers] were negatively worded and were reverse-coded before analysis. The [construct] score was computed as the mean of the [number] items, so that higher scores indicate higher [construct]. Scores were calculated only for respondents who answered at least [number] of the [number] items. Internal consistency was [acceptable / good], with Cronbach's alpha = [value]."
Common mistakes to avoid:
- Using Recode into Same Variables and losing the original data.
- Forgetting to reverse a negatively worded item, or reversing the same item twice.
- Including both the original and the reversed version of an item in the scale score.
- Leaving missing-value codes such as 99 undefined, so they enter the calculation and inflate the scale score.
Recoding and computing scale scores is a short step, but every later analysis depends on it. Start from a saved copy of your raw data, define missing values, and use the original scale instructions to identify reversed items and the scoring method. Use Recode into Different Variables so your originals stay unchanged, check every recode with a crosstab, and confirm reliability before you compute the score. Use Compute Variable with a clear rule for missing items, save your syntax, and report each decision in your methodology chapter. A careful, documented process makes your results easier to trust and easier to defend.