Rosenberg Self-Esteem Scale: scoring and interpretation

The Rosenberg Self-Esteem Scale (RSES) is, in its custodian’s words, perhaps the most widely used self-esteem measure in social science research, and one of the most carelessly cited. Search for “Rosenberg self-esteem scale interpretation” and the first page will hand you a three-band table — high, average, low — presented as though it were official. It is not. The University of Maryland Department of Sociology, which is custodian of the instrument, answers that exact question on its own site in one sentence: there are no discrete cut-off points to delineate high and low self-esteem.

What it measures, and why that matters

The custodian’s page defines the construct: self-esteem is a positive or negative orientation toward oneself, an overall evaluation of one’s worth or value. The scale was developed in the 1960s with 5,024 high school juniors and seniors from 10 randomly selected schools in New York State.

Self-esteem is not a DSM category; it appears inside other categories. In the DSM-IV-TR dysthymic disorder criteria reproduced in SAMHSA’s TIP 48, low self-esteem sits among the symptoms accompanying depressed mood — a criterion within a diagnosis, never a diagnosis. And it is simultaneously a legitimate treatment target: the meta-analysis of 77 longitudinal studies by Sowislo & Orth (2013) supports the vulnerability model, with low self-esteem predicting depression (β = −.16) more strongly than the reverse (β = −.08).

Structure and administration

Ten items — five positively worded, five negatively worded — with four options from strongly agree to strongly disagree. It was designed as a Guttman scale and is now almost always Likert-scored. It is self-administered: the PhenX Toolkit protocol specifies no specialized personnel or equipment and under 15 minutes; the source age range is 15 and older, though its Working Group recommends the measure down to age 10. The custodian reports test-retest correlations of .82 to .88 and alphas of .77 to .88.

The standard version sets no recall window: it asks about how the person generally feels, not about the last two weeks. That is a structural difference from the PHQ-9 and the GAD-7, and it explains why the RSES does not behave like them over time.

The items are in the public domain and the custodian publishes its own item sheet. We do not reproduce them here, for the reason set out in the guide to psychological test licensing. And one instruction almost nobody follows: do not print the asterisks marking the reverse-keyed items on the sheet you hand the client.

Scoring: two metrics and one trap

The official rule assigns 3, 2, 1 and 0 points (strongly agree through strongly disagree) to items 1, 2, 4, 6 and 7, and the reversed values to items 3, 5, 8, 9 and 10; the range runs 0 to 30. The same custodian allows the other convention: weights of 1 to 4 instead of 0 to 3, which moves the range to 10–40.

Convention	Weight per response	Total range	Where you will meet it
0–3	0 to 3	0–30	Custodian’s page; English literature
1–4	1 to 4	10–40	Most Spanish-language studies

Do not mix them, and never omit which one you used: a 24 means opposite things on each.

The column-numbering trap

The official sheet numbers its columns 1 = strongly agree, 2 = agree, 3 = disagree, 4 = strongly disagree. That numbering runs in the opposite direction from the weights the same page assigns to positively worded items, where strongly agree is worth 3, not 1. The column numbers are labels, not scoring weights. Anyone scoring off the columns inverts the whole scale and turns high self-esteem into low.

There is no cutoff, and the circulating bands have no source

The custodian points you to the literature for the population you care about. We looked for the origin of the two most widely circulated band tables and could not locate one. The English version — 15 to 25 within the normal range, below 15 indicating low self-esteem — is frequently attributed to the custodian itself and appears on none of its current pages. The Spanish version — 30–40 high, 26–29 average, under 25 low — appears only on document repositories, content sites and thesis appendices, with no attribution to any published study. Treat both as folklore: do not cite them and do not paste them into a note.

The contrast with instruments that do have cutoffs is the lesson. The PHQ-9 and the GAD-7 have thresholds because they were validated against an independent diagnostic interview, with published sensitivity and specificity for a specific disorder. The RSES was not built that way: we located no diagnostic-accuracy study for it, in any language, against any reference standard. There could hardly be one — there is no “self-esteem disorder” to serve as the criterion.

What does exist: percentiles, not cutoffs

The most useful population-specific work we located comes from Reynoso-González and colleagues (2022), with undergraduates at three public institutions in Coahuila, Jalisco and Nayarit, Mexico. A bifactor model with a general factor fit best, internal consistency was adequate (α = 0.87 total; 0.80 and 0.79 by dimension), and the paper publishes bands on the 10–40 metric: low 10–26, middle 27–35, high 36–40, with a mean of 30.62 and SD of 6.52.

Read them for what they are: percentiles — low, below the 25th; middle, 25th to 75th; high, above the 75th — computed on an incidental sample of undergraduates. They are descriptive bands for that sample, not clinical cutoffs and not a national norm. The paper’s own abstract calls them cut-off points; its method derives them from percentiles, not from a diagnostic standard. Another crack: the methods report 675 participants, a table 692. And norms are subgroup-specific: in the US normative data of Sinclair and colleagues (2010), scores varied significantly by age, education, employment, income, marital status and racial and ethnic group.

Translation is not validation

In English, the RSES debate is about dimensionality: Marsh, Scalas & Nagengast (2010) refute the unidimensional model and support one substantive factor plus response-style method factors stable over eight years; Gnambs & Schroeders (2020), with 12,437 students in a nationally representative German assessment, show that this wording effect shrinks as reading competence and reasoning increase. With adolescents and with low-literacy clients, the negatively worded items are the fragile part.

If you work with Spanish-speaking clients, the operative problem is different: several distinct Spanish texts circulate under the same name, and the custodian endorses none of them — it points you to the literature in the language you are using and says it is not a repository for translations. The four with published evidence behind them:

Spain, clinical sample. Vázquez Morejón and colleagues (2004), 533 community mental-health outpatients, Echeburúa’s 1995 translation: alpha 0.87, r = 0.72 at two months, r = 0.74 at one year, and r = −0.61 with the SCL-90-R total.
Chile. Rojas-Barahona, Zegers & Förster (2009), 473 adults in the Santiago Metropolitan Region: alpha 0.754 and two factors. This is the paper that states the 10–40 metric in Spanish.
Mexico, older adults. De León Ricardi & García Méndez (2016), 220 people aged 60 to 88 in Mexico City, built on the Chilean translation with Mexican linguistic adjustment: alpha 0.73.
Colombia, the necessary counterweight. Cogollo, Campo-Arias & Herazo (2015), a dichotomous version with a 30-day recall window, 951 schoolchildren aged 10 to 15: Kuder-Richardson 0.63, not unidimensional. Changing the response format changes the instrument.

Three absences worth stating out loud. We located no Mexican validation in a clinical sample — the Mexican studies we did locate are with undergraduates and with community-dwelling older adults. For Mexican adolescents we located no psychometric study we would cite. Nor did we locate Mexican test-retest data. A useful irony: the Mexican undergraduate study reports using a dichotomous Colombian adaptation while scoring it on a 1–4 Likert metric.

What a score does not tell you

It does not diagnose. There is no disorder to confirm or rule out with a self-esteem number.
It does not screen. With no reference standard and no published diagnostic accuracy, a low score detects nothing; it describes.
It does not compare across languages. In Schmitt & Allik (2005), with 16,998 participants across 53 nations and 28 languages, negatively worded items were interpreted differently across nations, and the authors conclude that direct cross-cultural comparison may have limited value.
It does not measure last week. The custodian warns that self-esteem is generally a stable characteristic of adults, not easily manipulated experimentally and unrealistic to think of as something taught. Re-administering weekly produces noise, not a progress curve: sensible intervals are measured in months, as the guide to measuring therapy progress sets out.
It says nothing about risk. The RSES contains no suicidal-ideation item. If any risk disclosure appears, an immediate clinical risk assessment under your own protocol and training is what follows. This guide will not hand you a step-by-step protocol — that comes from the professional guidelines that apply where you practice — and be wary of anyone who offers one in a blog post. Make sure your client leaves knowing the crisis line serving their location and local emergency services.

Licensing: public domain, and still cited

The current notice on the custodian’s page is unambiguous: the scale is now in the public domain, may be used without charge and without notifying the Sociology Department, and that permission extends to making translations or adaptations, consistent with traditional scholarly attribution practices.

The older FAQ text survives on the same site, from June 2021, framing use as permission granted by Rosenberg’s widow and asking for credit. The public-domain notice — undated itself, on pages last modified in September 2025 — controls, but the request for credit still makes sense, and the credit the custodian asks for is not the 1965 first edition but the revised one: Rosenberg, Morris. 1989. Society and the Adolescent Self-Image. Revised edition. Middletown, CT: Wesleyan University Press.

Where the score lives in the note

Many jurisdictions regulate what a progress note must contain; in Mexico, for example, the clinical-record standard requires the progress note to describe relevant results of previously requested diagnostic services, and requires every note to carry the date, time, full name of its author and a signature — the scope of that standard for psychological practice is covered in the guide to the Mexican clinical record standard. Verify the record-content rules that bind you where you practice. With an instrument that has no cutoff, the entry has to carry the context the number does not:

“RSES (Rosenberg, 1989; Rojas-Barahona et al. 2009 Spanish translation), administered 03/12/2026: 24/40. No established clinical cutoff; used as a monitoring baseline, not as a diagnostic criterion.”

Writing “low self-esteem (24)” with no metric, no version and no date turns a measurement into a label, and it is one of those clinical documentation mistakes that cannot be reconstructed later. The absence of a cutoff forces you to write the objective around change, not a threshold: “an increase of at least 5 points on the RSES, 10–40 metric, sustained across two administrations three months apart” is evaluable; “improve self-esteem” is not. That mechanic is covered in the guide to SMART treatment goals.

Recording version, metric, date and raw score — and being able to compare today’s measurement with one from six months ago — is easy with one client and heavy with thirty. That is where a platform helps: gesell.ai includes built-in validated scales with automatic scoring, PHQ-9 and GAD-7 among them, whose scores land in the client’s structured chart ready for your note; the RSES you still score and contextualize yourself. The instrument contributes the data point; the clinical reading stays yours, and the output is decision support you evaluate.