Home / Resources / Psychological Measures / SF-36
36-Item Short Form Health Survey:
Scoring and Domain Profiles
The 36-Item Short Form Health Survey (SF-36) profiles health-related quality of life across eight physical and mental-health domains, each recoded to a 0 to 100 scale (Ware & Sherbourne, 1992). There is no 36-item sum, and the scoring version decides the numbers. Here is how SF-36 scoring works under RAND-36 version 1.0, with SPSS and R syntax and a methods paragraph that names the version.
What the SF-36 Measures
The 36-Item Short Form Health Survey (SF-36) was developed within the Medical Outcomes Study to profile health-related quality of life across eight physical and mental-health domains (Ware & Sherbourne, 1992). It is a generic measure rather than a psychiatric symptom scale, useful for patient-reported outcomes in clinical trials, health services research and population surveys.
A critical distinction is the scoring version. RAND’s public 36-Item Health Survey 1.0 uses linear 0 to 100 recoding and averages items within each scale (Hays et al., 1993). RAND explicitly notes that its scoring differs from the MOS SF-36 procedure distributed elsewhere. The code below implements RAND-36 1.0, not a generic algorithm for every SF-36 version.
Structure of the SF-36
- Items: 36 items; item 2 is the health-transition item and is excluded from the eight domain composites.
- Domain mapping: physical functioning, items 3 to 12; role physical, 13 to 16; role emotional, 17 to 19; energy/fatigue, 23, 27, 29 and 31; emotional well-being, 24, 25, 26, 28 and 30; social functioning, 20 and 32; pain, 21 and 22; general health, 1, 33, 34, 35 and 36.
- Response scale: mixed response sets of 2, 3, 5 or 6 categories depending on the item, with recall mainly of present health and the preceding four weeks.
- Recoding: responses need item-specific recoding to 0 to 100, with higher meaning more favorable health. There is no single list of reverse-scored items equivalent to a uniform Likert reversal.
- Missing data: RAND uses the available-item mean within each domain; if all items in a domain are missing, its score is missing.
- Total: standard domain scores are not a simple sum of all 36 item responses. Mental and physical component-summary algorithms are version-specific and not implemented here.
- Versions: RAND-36 1.0 and MOS SF-36 scoring exist, and the SF-36v2 and acute forms require separate documentation.
The instrument itself is available from the RAND official 36-Item Short Form Survey scoring page.
How to Score the SF-36
- Verify that sf1 to sf36 are the original RAND-36 version 1.0 precoded category numbers and that item order has not been changed.
- Apply RAND’s item-specific six recoding mappings exactly as implemented below. Do not apply the same reverse-key formula to all 36 items.
- Average the recoded 0 to 100 scores in each of the eight domain groups using the available answered items, as RAND instructs. A domain with no answered item remains missing.
- Retain item 2 only for the health-transition indicator; do not include it in the eight domain scores.
- Do not label the results SF-36v2 norm-based component T-scores; those require a distinct, version-specific scoring key and reference norms.
* Illustrative syntax: sf1 to sf36 must hold the original RAND-36 v1.0 category numbers.
RECODE sf1 sf20 sf22 sf34 sf36 (1=100) (2=75) (3=50) (4=25) (5=0) INTO sf_s1 sf_s20 sf_s22 sf_s34 sf_s36.
RECODE sf3 sf4 sf5 sf6 sf7 sf8 sf9 sf10 sf11 sf12 (1=0) (2=50) (3=100) INTO sf_s3 sf_s4 sf_s5 sf_s6 sf_s7 sf_s8 sf_s9 sf_s10 sf_s11 sf_s12.
RECODE sf13 sf14 sf15 sf16 sf17 sf18 sf19 (1=0) (2=100) INTO sf_s13 sf_s14 sf_s15 sf_s16 sf_s17 sf_s18 sf_s19.
RECODE sf21 sf23 sf26 sf27 sf30 (1=100) (2=80) (3=60) (4=40) (5=20) (6=0) INTO sf_s21 sf_s23 sf_s26 sf_s27 sf_s30.
RECODE sf24 sf25 sf28 sf29 sf31 (1=0) (2=20) (3=40) (4=60) (5=80) (6=100) INTO sf_s24 sf_s25 sf_s28 sf_s29 sf_s31.
RECODE sf32 sf33 sf35 (1=0) (2=25) (3=50) (4=75) (5=100) INTO sf_s32 sf_s33 sf_s35.
COMPUTE rand36_physical_functioning=MEAN(sf_s3,sf_s4,sf_s5,sf_s6,sf_s7,sf_s8,sf_s9,sf_s10,sf_s11,sf_s12).
COMPUTE rand36_role_physical=MEAN(sf_s13,sf_s14,sf_s15,sf_s16).
COMPUTE rand36_role_emotional=MEAN(sf_s17,sf_s18,sf_s19).
COMPUTE rand36_energy_fatigue=MEAN(sf_s23,sf_s27,sf_s29,sf_s31).
COMPUTE rand36_emotional_wellbeing=MEAN(sf_s24,sf_s25,sf_s26,sf_s28,sf_s30).
COMPUTE rand36_social_functioning=MEAN(sf_s20,sf_s32).
COMPUTE rand36_pain=MEAN(sf_s21,sf_s22).
COMPUTE rand36_general_health=MEAN(sf_s1,sf_s33,sf_s34,sf_s35,sf_s36).
EXECUTE.# Variables sf1 ... sf36 hold pre-coded questionnaire categories.
# RAND 36 v1.0 scoring key; item 2 is health transition, not in 8 scales.
maps <- list(
`1,20,22,34,36`=c(100,75,50,25,0),
`3,4,5,6,7,8,9,10,11,12`=c(0,50,100),
`13,14,15,16,17,18,19`=c(0,100),
`21,23,26,27,30`=c(100,80,60,40,20,0),
`24,25,28,29,31`=c(0,20,40,60,80,100),
`32,33,35`=c(0,25,50,75,100))
for (grp in names(maps)) {
key <- maps[[grp]]
for (i in as.integer(strsplit(grp,",")[[1]])) {
x <- data[[paste0("sf",i)]]
data[[paste0("sf_s",i)]] <- key[match(x,seq_along(key))]
}
}
scales <- list(
physical_functioning=c(3,4,5,6,7,8,9,10,11,12),
role_physical=c(13,14,15,16),
role_emotional=c(17,18,19),
energy_fatigue=c(23,27,29,31),
emotional_wellbeing=c(24,25,26,28,30),
social_functioning=c(20,32),
pain=c(21,22),
general_health=c(1,33,34,35,36))
for (nm in names(scales)) {
x <- data[paste0("sf_s",scales[[nm]])]
z <- rowMeans(x,na.rm=TRUE) # RAND: available-item mean
z[is.nan(z)] <- NA_real_
data[[paste0("rand36_",nm)]] <- z
}Interpreting SF-36 Scores
| Score band | Interpretation |
|---|---|
| 0 | Worst possible reported health state on a domain |
| 0 to 100 for each of eight domain scales | Continuous RAND-36 health-status scale |
| 100 | Best possible health state on a domain |
| Universal good/poor clinical cut-offs | Not defined by RAND’s scoring instructions |
Source: RAND (n.d.), 36-Item Short Form Survey (SF-36) Scoring Instructions.
Higher scores indicate better self-reported health on all eight domains. Interpretation should use domain profiles and population-specific reference data where appropriate, not an unweighted sum of domain scores as though it were an official ninth scale.
Version warning: RAND-36 scoring is not interchangeable with MOS SF-36 or SF-36v2 norm-based algorithms. Confirm the administered questionnaire and item coding before using the included SPSS or R code; otherwise computed scores may be systematically wrong.
Psychometric Properties
| Property | Published finding | Source |
|---|---|---|
| Internal consistency, physical functioning | α = .93 in RAND-reported Medical Outcomes Study data | RAND scoring instructions, Table 3 |
| Internal consistency, emotional well-being | α = .90 in the same sample | RAND scoring instructions, Table 3 |
| Internal consistency, eight domains | Reported α range .78 to .93 across the eight RAND domains in the Medical Outcomes Study data | RAND scoring instructions, Table 3 |
| Content validity | The original SF-36 content represents eight physical and mental health concepts | Ware & Sherbourne (1992) |
| Diagnostic classification | Not a stand-alone diagnostic screener; domain changes require design-specific interpretation | Ware & Sherbourne (1992) |
These are source-specific results and cannot be assumed to hold as reliability or accuracy estimates in a new sample.
Using the SF-36 in a Study
Predefine the primary RAND-36 domain or domains from the research hypotheses. Correlated domains should not automatically be replaced with an arbitrary overall SF-36 total. Each prespecified domain can be modeled with Multiple Linear Regression, or with Linear Mixed Models (Multilevel Modeling) when patients are followed over time, with attention to ceiling effects in the role limitation domains and to multiple testing across eight endpoints.
Report whether you used RAND-36 1.0 or another SF-36 scoring method, because this changes item handling and potentially the reported values. For sample-specific reliability, calculate alpha or omega for the multi-item domains through Reliability Analysis (Cronbach’s Alpha), and summarize the two-item scales by their inter-item correlation instead. Describe how incomplete responses were treated, and consider Missing Data Analysis when whole forms are absent. A universal minimal important difference across all eight domains and patient groups is not specified by the original framework.
Related field: Health Sciences
Reporting the SF-36
Health-related quality of life was assessed using the 36-item Medical Outcomes Study health survey (Ware & Sherbourne, 1992), scored according to the RAND 36-Item Health Survey 1.0 instructions. Items were recoded to 0 to 100 and averaged into eight domain scores, with higher values indicating better health. Domain missing values were handled using RAND’s available-item averaging procedure. Reliability for the prespecified domain(s) was [α = XX], and the primary outcome was [domain].
References
Hays, R. D., Sherbourne, C. D., & Mazel, R. M. (1993). The RAND 36-Item Health Survey 1.0. Health Economics, 2(3), 217-227. https://doi.org/10.1002/hec.4730020305
RAND. (n.d.). 36-Item Short Form Survey (SF-36) scoring instructions. https://www.rand.org/health/surveys/mos/36-item-short-form/scoring.html
Ware, J. E., Jr., & Sherbourne, C. D. (1992). The MOS 36-item short-form health survey (SF-36): I. Conceptual framework and item selection. Medical Care, 30(6), 473-483. https://doi.org/10.1097/00005650-199206000-00002
Frequently Asked Questions
Can I sum all 36 SF-36 item responses?
No. RAND-36 defines eight domain scores after item-specific recoding, with one separate health-transition item.
Are RAND-36 and SF-36v2 scores identical?
No. RAND documents important differences between its 1.0 method and alternative SF-36 scoring procedures, so state which one you used.
How do I handle a missing response in a domain?
RAND calculates the mean of the available recoded items for that domain; an entirely unanswered domain remains missing.
Are SF-36 scores diagnostic?
No. They summarize dimensions of perceived health-related quality of life and functioning.
NEXT STEP
Need SF-36 Domain Scores
Built and Verified?
OnlineSPSS is a statistical consultancy for researchers. Share your design and dataset, and a statistically trained Research Coordinator reviews it. The scope and one fixed project fee are confirmed before work begins.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014