The PHQ-9 was presented as part of a large survey conducted by Leipzig University between December 2020 and March 2021.

The goals of the survey were (a) to assess prevalence rates of a variety of relevant physical or mental disorders and related risk behaviors (descriptive epidemiology), (b) to examine causes and conditions of these disorders (analytic epidemiology), and (c) to analyze psychometric properties and provide German population norms for clinical-psychological instruments. The survey was carried out by the contractor USUMA Markt- und Sozialforschung an independent institute for opinion and social research.

It consisted of two parts. The first part was guided by a trained interviewer and collected extensive demographic as well as household information. Survey contents in this part were based on principles of the German Statistisches Bundesamt (Federal Statistical Office).

The second part consisted of paper-based self-administered questionnaires which the participants filled in independently. Interviewers remained out of view but available for questions. Prior to their participation in the survey all participants obtained a written copy of the confidentiality agreement providing details regarding the handling of their personal data. The study followed the Declaration of Helsinki. Minimum age for participation was 16 years. All participants provided informed consent prior to the interview. For under-aged participants at least one legal guardian was informed about the sampling procedure and the survey contents. All procedures were approved by the Ethics Committee of the Medical Faculty of the University of Leipzig (Az.: 474/20-ek).

As Germany does not keep a central population registry, representativeness of the sample was ensured by using the ADM sampling system F2F. This sampling procedure consists of three steps. In a first step, the area of the Federal Republic of Germany is divided into regions of which 258 are sampled with sampling probability proportional to the number of households. In a second step, 5676 households are selected based on a random route procedure. Finally, the target person within each household is identified using a Kish selection grid (24). Further details regarding the sampling procedure, COVID measures and sample representativeness can be found in the Supplementary Materials (section A). Details regarding response can be obtained from Figure 1 . The following analyses are based on data from N = 2519 participants which corresponds to a response rate of 42%. Figure 1 presents a flowchart outlining the sampling procedure and reasons for non-response. Table 1 provides sample descriptives.

As the survey served multiple epidemiological purposes, only those measures that were used in the validation process are discussed in this paper. In addition to extensive demographic information (see Table 1 ), health related behavior, such as the number of sick days, doctor visits, and hospital stays, were assessed. The following measures were used for the validation of the scale at hand.

Supplementary Table C5 (in the Supplementary Materials ) displays means and standard deviations for the nine items of the PHQ-9 in the total sample as well as effect sizes for mean differences regarding gender. On the item-level there was a consistent pattern of female participants exhibiting higher mean depression scores as well as higher variability on most PHQ-9 items. Effect sizes (Cohen’s d) of these mean differences were very small and ranged from d = -0.05 95% CI [-0.13,0.03] to d = 0 95% CI [-0.08,0.08].

To determine evidence of construct validity of the PHQ-9, correlation coefficients were calculated with related instruments. In line with our hypotheses, there were high positive correlations between the PHQ-9 and measures of somatization, anxiety and depression as assessed by BSI-18 subscales (see Supplementary Table C7 in the Supplementary Materials). In the same vein, the GAD-7 assessing anxiety showed positive correlations with the PHQ-9.

Table 2 shows cumulative percentiles of PHQ-9 scores for the total sample. Additional norms split by gender as well as age group can be found in the Supplementary Material (see Supplementary Tables C8 , C9 ). Table 3 reports absolute and relative frequencies per severity category. We neither endorse nor have verified this classification but provide it as a mere descriptive to facilitate comparing results across studies.

Cronbach’s alpha of the PHQ-9 for the full sample was α = 0.90, 95% CI [0.89, 0.91]. McDonald’s omega of the PHQ-9 for the full sample was ω = 0.93, 95% CI [0.92, 0.94].

A CFA was conducted to assess the unidimensional structure of the PHQ-9. The fit indices indicated reasonable model fit, with a robust CFI of 0.91, a robust TLI of 0.89, and an SRMR of 0.044. However, the robust RMSEA was 0.17 (90% CI [0.153, 0.186]), suggesting some misfit.

To improve the model, we inspected modification indices and introduced residual correlations between items with overlapping content: #1 (Little interest or pleasure) with #2 (Down, depressed, hopeless) and #3 (Sleep problems) with #4 (Tired, little energy). These adjustments improved the fit indices (CFI = 0.96, RMSEA = 0.11), as detailed in Supplementary Table C2 .

Despite these modifications, factor score correlations between the original and adjusted models remained high (r > 0.999), confirming the stability of the latent structure and supporting the unidimensionality of the PHQ-9. Strong standardized factor loadings (0.79–0.89) further reinforced this. A SEM path diagram can be found in Supplementary Figure C2 of the Supplementary Materials .

The fit measures obtained in the measurement invariance analyses of the PHQ-9 are presented in Supplementary Table C4 in the Supplementary Materials . Adequate CFI and RMSEA differences were found for all invariance steps and groups. The sensitivity analyses using MLR estimation confirmed measurement invariance regarding age, gender and education. Given the very small CFA and RMSEA differences in the original analysis using WLSMV estimation, measurement invariance regarding age × gender and income is likely, yet inconclusive.

Despite the current study being based on high quality data from a large representative sample it is not without limitations. First, the response rate was only 42.6%. However, general population studies commonly have significantly lower response rates than clinical studies and the response rate of this study is comparable to similar surveys [e.g., (31, 43, 44)]. Despite significant efforts to maximize sample representativeness, some degree of non-response remains unavoidable with the current design and there is a potential for bias due to non-response. Unfortunately, the possibility of non-response bias cannot be systematically assessed as no demographic information of non-responders is available. This would only be possible if sampling were based on registry data which is not accessible without government permit in Germany. Furthermore, the current study does not allow any conclusions regarding the diagnostic efficiency of the PHQ-9 as no clinical interviews were conducted. While the presented norms and psychometric properties can serve as valuable reference data for clinical research the generalizability to clinical samples is limited.

In summary, the German version of the PHQ-9 has shown sound psychometric properties in a large representative population sample. While the percentiles for the sum-score are comparable to similar studies in community samples, the change in symptom burden compared to previous German norm values from over a decade earlier make the present study an important reference. Furthermore, the results from the present study serve as a notable counter point in the interpretation of COVID-19 related findings including the PHQ-9. We suggest updating the norms again in the near future to gain a deeper understanding whether the present findings have to be interpreted as pandemic related or as “the new normal”.