To reverse-code a Likert item, compute new score = (scale maximum + scale minimum) − old score; on a 1-to-5 scale that is 6 − old, so 1 becomes 5, 2 becomes 4, 3 stays 3. Do it before computing scale scores, Cronbach's alpha or factor analysis. Afterwards, check that every item in the scale has a positive Corrected Item-Total Correlation, ideally .30 or higher (Field, 2018); a negative value almost always means an item was not recoded, or was recoded twice.
Why scales include reversed items
Questionnaire designers mix positively and negatively worded items so that respondents who agree with everything (acquiescence) or tick the same column throughout can be detected. "I enjoy my job" and "I often think about quitting" both measure job satisfaction, but in opposite directions. Before analysis, all items must point the same way, so that a high score always means more of the construct. Reversed items have a cost, too: Weijters and Baumgartner (2012) show that respondents often misread them, which weakens reliability and factor structure. That is why recoding and checking matter.
The formula for any scale
| Scale | Formula | Mapping |
|---|---|---|
| 1–5 | 6 − x | 1→5, 2→4, 3→3, 4→2, 5→1 |
| 1–7 | 8 − x | 1→7, 2→6, 3→5, 4→4 … |
| 0–4 | 4 − x | 0→4, 1→3, 2→2 … |
| 0–10 | 10 − x | 0→10, 3→7, 10→0 |
The general rule is (max + min) − x. A common mistake is to use max − x on a scale that starts at 1, which turns a 1–5 scale into 0–4 and shifts the reversed item's mean down by one point.
Worked example
A four-item Job Satisfaction scale is answered on a 1–5 scale. JS3, "I often think about quitting this job", is the reversed item. Respondent 17 answers JS1 = 4, JS2 = 5, JS3 = 2, JS4 = 4.
Recoded JS3 = 6 − 2 = 4. The respondent's scale mean is (4 + 5 + 4 + 4) / 4 = 4.25. Without recoding, it would have been (4 + 5 + 2 + 4) / 4 = 3.75, understating satisfaction.
At the sample level, before recoding, the reliability output shows JS3 with a Corrected Item-Total Correlation of −.41 and Cronbach's alpha of .52. After recoding into a new variable JS3_R, the item-total correlation is .58 and alpha rises to .81, above the .70 threshold (Nunnally, 1978). The negative sign was the clue.
Good practice
- Create a new variable (JS3_R) instead of overwriting the original, so the step can be checked and repeated.
- Use the scale documentation to identify reversed items. Validated scales list them; do not guess from wording alone.
- Recode once. Recoding twice restores the original direction. Keep a log.
- Check after recoding: item-total correlations should all be positive, and EFA loadings should share the same sign.
- Watch missing-value codes: if missing answers are coded 99 or 0, exclude them before applying the formula.
Reverse-coded items and factor analysis
Even after correct recoding, reversed items sometimes form a separate factor in EFA, made up only of the negatively worded items. This is usually a wording artefact rather than a new construct (Weijters & Baumgartner, 2012). If it happens, check whether the "factor" makes theoretical sense; if it does not, report it as a method effect and consider whether the reversed items should stay. Never delete them silently.
Doing it in common software
In SPSS: Transform, Compute Variable, JS3_R = 6 − JS3. In Excel or Google Sheets: =6-C2 in a new column. In R: data$JS3_R <- 6 - data$JS3. In JASP or jamovi: compute a new variable with the same expression. The logic is identical everywhere.
Reporting it
Item JS3 was negatively worded and was reverse-coded (6 − x) before analysis so that higher scores indicated greater job satisfaction. All Corrected Item-Total Correlations were above .30 (Field, 2018), and Cronbach's alpha was .81.
Run it free
The free survey data cleaning tool reverse-codes selected columns with the correct formula for your scale range, flags straight-lining and keeps a log of every change. Then check reliability with the Cronbach's alpha calculator.
FAQ
Do I recode before or after EFA?
Before. Unrecoded items load negatively and can distort the factor solution and KMO.
Should I remove reversed items if they perform badly?
Only after checking recoding and wording. If a correctly recoded item still has a very low item-total correlation, removal may be justified; report it.
Is the midpoint changed by reverse coding?
No. On an odd-numbered scale the midpoint stays the same (3 on a 1–5 scale).
References
- Field, A. (2018). Discovering statistics using IBM SPSS Statistics (5th ed.). Sage.
- Nunnally, J. C. (1978). Psychometric theory (2nd ed.). McGraw-Hill.
- Weijters, B., & Baumgartner, H. (2012). Misresponse to reversed and negated items in surveys: A review. Journal of Marketing Research, 49(5), 737–747. https://doi.org/10.1509/jmr.11.0368