Depression Rating Scales Compared: HAM-D, MADRS, PHQ-9

“Which depression rating scale should you use?” has no one-size answer, and pretending otherwise is how two studies measure the “same” thing and never line up. That is why it helps to see the main depression rating scales compared directly—the HAM-D, the MADRS, the PHQ-9, and Gottschalk-Gleser content analysis. The real choice resolves into four decisions: who does the rating, whether you need screening, severity, or change, how much time you have, and which slice of depression you actually care about.

Get those four right and the scale nearly picks itself. Get them wrong and you can rate the same patient on two scales and reach different conclusions—which, as the evidence below shows, happens more often than the field admits.

Depression rating scales compared: the four-way matrix

The four instruments below measure the same broad construct—depressive severity—but were built for different jobs, and those design choices decide which one fits your study. The matrix has the depression rating scales compared side by side on the axes that actually drive the decision: who rates them, how many items, how long they take, what job they do best, and how sensitive they are to change.

Axis	PHQ-9	MADRS	HAM-D (HDRS-17)	Gottschalk-Gleser
Rater	Self-report	Clinician	Clinician	Trained coder (from speech)
Items	9	10	17 (scored)	7 subscales, clause-level
Score range	0–27	0–60	0–52	Per-subscale magnitude scores
Time to administer	~2–5 min	~15–20 min	~15–20 min	5-min speech sample + intensive coding
Primary job	Screening + self-monitoring	Detecting change	Severity; legacy endpoint	Measuring affect in speech
Change-sensitivity	Moderate	High (by design)	High	Research use
Somatic loading	Moderate (DSM-mapped)	Low (2 of 10)	High (3 insomnia items)	One somatic subscale
Original source	Kroenke et al., 2001	Montgomery & Åsberg, 1979	Hamilton, 1960	Gottschalk & Hoigaard-Martin, 1986

Each scale also has a full explainer of its own: the PHQ-9 scoring guide, the MADRS scoring guide, the Hamilton scale guide, and the Gottschalk-Gleser content-analysis guide. This page is the hub that decides between them.

Which “depression” is each scale measuring?

Each of these scales measures a slightly different depression, not the same one at different resolutions. When Eiko Fried disaggregated seven common depression scales into their component symptoms, the 125 items resolved into 52 distinct symptoms, and the content overlap between scales was weak—a mean Jaccard index of 0.36, where 1.0 would mean identical content (Fried, 2017).

The breakdown is stark. Of those 52 symptoms, 40% appear in only a single scale, and only one—sad mood—is captured specifically by every scale (Fried, 2017). The MADRS, deliberately lean and mood-focused, has one of the lowest content overlaps with the others (0.30), which is a feature of its design rather than a flaw.

This is the thesis that should govern every comparison that follows. Because the instruments sample different symptom sets, a treatment that moves one scale need not move another, and a finding obtained with one scale may not replicate with a second (Fried, 2017). “Depression severity” is not one quantity measured with error; it is several overlapping constructs wearing the same name. Choosing a scale is really choosing which depression you will see—which is why any honest look at the best depression scale research starts with content, not just reliability.

Clinician-rated or self-report? (axis 1: who does the rating)

The first fork is who assigns the score. The PHQ-9 is self-report—the patient rates themselves—while the HAM-D and MADRS are clinician-rated, scored by a trained interviewer from what the patient says and shows. Gottschalk-Gleser sits outside the split: it is neither, scoring a transcript of the patient’s own speech against defined categories.

The trade-off is real in both directions. Self-report is fast, cheap, scalable, and captures the patient’s inside view, but it is sensitive to insight, mood on the day, and response style. Clinician rating adds structured observation—the MADRS even reserves one item, apparent sadness, for dejection the interviewer sees rather than what the patient reports—at the cost of rater time and rater drift, where two clinicians score the same interview differently.

That difference shapes where each scale is used. In a review of 2,632 registered depression trials, clinician-administered scales dominated as inclusion and remission criteria in drug trials, while self-report questionnaires were primarily used in behavioral trials, and the gap widened over two decades (von Glischinski et al., 2021). The clinician-rated vs self-report depression question is covered in depth in our companion post on clinician-rated versus self-report scales; for the self-report end specifically, the PHQ-9 is the reference implementation.

Screening, severity, or change? (axis 2: the job the score does)

The second axis is what the number is for. The PHQ-9 is optimized to screen: at its conventional cutoff of ≥10 it caught major depression with a pooled sensitivity of 0.88 and specificity of 0.85 across 58 studies (Levis, Benedetti & Thombs, 2019). The HAM-D and MADRS are built to grade severity and, above all, to track change—the MADRS explicitly so, having kept the 10 symptoms that moved most with antidepressant treatment (Montgomery & Åsberg, 1979).

Between the two clinician scales, the change signal is closer than reputations suggest. Head-to-head, treatment effect sizes were 0.49 for the MADRS and 0.53 for the HAM-D (Khan et al., 2002), and across 161 comparisons in 80 trials neither scale was systematically more sensitive (Guizzaro et al., 2020). The honest caveat from that same analysis: the two are interchangeable in meta-analyses but the same trial can read differently on each, especially in small studies.

So when you pick a depression outcome measure, decide the job first. Screening routes to the PHQ-9; a clean change endpoint favors the MADRS; continuity with the vast older literature favors the HAM-D. We compare the two clinician scales in detail in MADRS vs HAM-D, and the full item sets live on the MADRS and Hamilton scale pages.

Brief or comprehensive? (axis 3: administration cost)

The third axis is time and training, and it separates the scales as cleanly as any psychometric property. The PHQ-9 is self-administered in roughly two to five minutes and needs no rater, which is why primary care leans on it. The MADRS and HAM-D each need a clinician interview of about fifteen to twenty minutes, plus the standardization that keeps raters aligned—structured guides such as the SIGMA for the MADRS and the SIGH-D for the HAM-D exist precisely to hold that reliability across interviewers (Williams, 1988; Williams & Kobak, 2008).

Gottschalk-Gleser is the most demanding of the four. It takes only a five-minute speech sample to collect, but the scoring is labor-intensive: trained coders are held to an inter-scorer reliability of 0.80 or better, and reaching that bar takes practice against previously scored samples (Gottschalk & Gleser, 1969). It buys something the checklists cannot—affect measured in the person’s own words—but it is a research instrument, not a bedside quick-screen.

Where does content analysis fit? (the qualitative outlier)

Gottschalk-Gleser content analysis is the outlier of the four: instead of a checklist filled in after an interview, it scores a transcript clause by clause across seven depressive subscales, weighting each clause by how strongly it signals the state (Gottschalk & Hoigaard-Martin, 1986). Where the PHQ-9, MADRS, and HAM-D produce one severity total, it produces a profile—hopelessness, self-accusation, psychomotor retardation, somatic concerns, death and mutilation, separation, and hostility outward—normalized for how much the person spoke.

That design makes it the qualitative counterweight in this comparison. It reads how someone talks about their distress rather than asking them to rate it, which is exactly what a symptom count throws away. It is closest in spirit to modern span-level text annotation, and it earns its place when the question is “what affect is in this speech,” not “how severe, on a fixed scale.” The full subscale structure is on the Gottschalk-Gleser depression scale page.

Which depression scale should you use?

Pick the instrument for the question, not its reputation. Choosing a depression scale comes down to matching one of four jobs to the tool built for it—the decision table below routes each goal to a scale and the reason behind it.

If your goal is…	Use	Because
Screen for depression or self-monitor between visits	PHQ-9	9 self-report items; ≥10 is the validated positive-screen cutoff (Levis et al., 2019)
Detect treatment-related change in a trial	MADRS	10 mood-weighted items built to be sensitive to change (Montgomery & Åsberg, 1979)
Stay comparable with decades of antidepressant data	HAM-D-17	The legacy endpoint; older results are denominated in HAM-D points (Hamilton, 1960)
Measure affect in a person's own words	Gottschalk-Gleser	Weighted, clause-level content analysis of speech (Gottschalk & Hoigaard-Martin, 1986)

Two routing notes keep this honest. First, the PHQ-9 vs HAMD choice is not self-report-versus-clinician alone—it is screening versus severity, and using a screen as a change endpoint (or a clinician severity scale as a mass screen) fights each tool’s design. Second, none of these totals convert one-to-one; when you must bridge scales, translate through percentage change from baseline rather than raw points, and never treat a MADRS total and a HAM-D total as the same currency.

How coding from a transcript unifies the four

Whichever scale you choose, the sane way to score it from a recorded interview is the same: code the passage where the subject reports each symptom, then rate that evidence on the scale’s own anchors, rather than filling a form from memory. This produces an auditable score—every rating points back to the exact words that justify it—and it is the workflow Tagaroo is built for.

Consider one short synthetic exchange, and watch how the same words feed different scales:

Interviewer: How have things been this past couple of weeks?

Subject: I can’t sleep, I’ve stopped bothering with anything, and honestly I keep thinking everyone would be better off without me.

On the PHQ-9, that reply loads the sleep, anhedonia, and item-9 (suicidal ideation) items. On the MADRS, it moves reduced sleep, inability to feel, and suicidal thoughts. On the HAM-D, it touches insomnia, work and activities, and the suicide item. In Gottschalk-Gleser terms, “everyone would be better off without me” is a weight-4 self-accusation clause—the single fragment you never want a total to bury.

Coding the evidence span rather than a global impression is what pays off across all four. It lets a reviewer check each rating, lets you compute inter-rater reliability across coders, and surfaces an endorsed suicide item on its own, independent of any total.

On data handling: transcript coding means working with sensitive language, so the sane default is de-identified text and a privacy-first setup. Tagaroo supports a browser-side anonymous mode so transcript content can stay local rather than being uploaded—worth checking against your ethics approval before any real interview data touches a tool.

Common mistakes when comparing depression scales

The costliest mistake is treating the scales as interchangeable measures of one thing. A few others recur:

• Assuming the totals are comparable. A PHQ-9 of 14, a MADRS of 26, and a HAM-D of 20 are not “the same” severity; the scales differ in range, content, and rater. Bridge through percentage change, not raw points.
• Using a screen as a change endpoint. The PHQ-9 is validated for screening at ≥10 (Levis et al., 2019); a clinician severity scale is the better instrument for tracking treatment change.
• Reading one scale’s result as depression in general. Because content overlap is weak, a finding from one scale may not replicate on another (Fried, 2017)—report which scale produced it.
• Ignoring the rater axis. Self-report and clinician-rated scales can diverge on the same patient; pooling them into one change curve mixes two vantage points (von Glischinski et al., 2021).

None of these are reasons to distrust any of the four. They are reasons to report a depression score with its context: the scale, the version, the cutoff, the rater, and whether the number came from a form or from evidence you can point to.

The practical upshot of comparing these depression rating scales side by side: there is no single best scale in the abstract, only a best scale for a defined job. Screen with the PHQ-9, track change with the MADRS, keep legacy comparability with the HAM-D, and measure affect in speech with Gottschalk-Gleser—and remember that each one shows you a different depression. If you code depressive symptoms from interviews, Tagaroo turns the PHQ-9, MADRS, Hamilton scale, and Gottschalk-Gleser scale into guided, evidence-anchored annotation workflows with inter-rater reliability computed as your coders work.

References

• Hamilton, M. (1960). A rating scale for depression. Journal of Neurology, Neurosurgery & Psychiatry, 23(1), 56–62. doi:10.1136/jnnp.23.1.56
• Montgomery, S. A., & Åsberg, M. (1979). A new depression scale designed to be sensitive to change. British Journal of Psychiatry, 134(4), 382–389. doi:10.1192/bjp.134.4.382
• Kroenke, K., Spitzer, R. L., & Williams, J. B. W. (2001). The PHQ-9: validity of a brief depression severity measure. Journal of General Internal Medicine, 16(9), 606–613. doi:10.1046/j.1525-1497.2001.016009606.x
• Gottschalk, L. A., & Hoigaard-Martin, J. (1986). A depression scale applicable to verbal samples. Psychiatry Research, 17(3), 213–227. (Building on Gottschalk, Winget & Gleser, 1969.)
• Gottschalk, L. A., & Gleser, G. C. (1969). The Measurement of Psychological States Through the Content Analysis of Verbal Behavior. University of California Press. doi:10.1525/9780520376762
• Fried, E. I. (2017). The 52 symptoms of major depression: lack of content overlap among seven common depression scales. Journal of Affective Disorders, 208, 191–197. doi:10.1016/j.jad.2016.10.019
• Khan, A., Khan, S. R., Shankles, E. B., & Polissar, N. L. (2002). Relative sensitivity of the Montgomery-Asberg Depression Rating Scale, the Hamilton Depression Rating Scale and the CGI in antidepressant clinical trials. International Clinical Psychopharmacology, 17(6), 281–285. doi:10.1097/00004850-200211000-00003
• Levis, B., Benedetti, A., & Thombs, B. D. (2019). Accuracy of Patient Health Questionnaire-9 (PHQ-9) for screening to detect major depression: individual participant data meta-analysis. BMJ, 365, l1476. doi:10.1136/bmj.l1476
• von Glischinski, M., von Brachel, R., Thiele, C., & Hirschfeld, G. (2021). Not sad enough for a depression trial? A systematic review of depression measures and cut points in clinical trial registrations. Journal of Affective Disorders, 292, 36–44. doi:10.1016/j.jad.2021.05.041
• Guizzaro, L., Morgan, D. D. V., Falco, A., & Gallo, C. (2020). Hamilton scale and MADRS are interchangeable in meta-analyses but can disagree at trial level. Journal of Clinical Epidemiology, 124, 106–117. doi:10.1016/j.jclinepi.2020.04.022
• Williams, J. B. W. (1988). A structured interview guide for the Hamilton Depression Rating Scale (SIGH-D). Archives of General Psychiatry, 45(8), 742–747. doi:10.1001/archpsyc.1988.01800320058007
• Williams, J. B. W., & Kobak, K. A. (2008). Development and reliability of a structured interview guide for the MADRS (SIGMA). British Journal of Psychiatry, 192(1), 52–58. doi:10.1192/bjp.bp.106.032532
• U.S. Food and Drug Administration. Major Depressive Disorder: Developing Drugs for Treatment (draft guidance for industry). FDA guidance document