Instrument development was guided by the PROMIS standards (Cella et al., 2007). Item generation involved drawing items from existing questionnaires in the three domains, standardizing them to the same response format and making small changes to wording where necessary. Item reduction was achieved by conducting an Exploratory Factor Analysis (EFA) on responses from a predominantly student sample to this draft set of items. Preliminary investigation of external validity was conducted via associations with self-reported physical and mental health outcomes.

To develop a comprehensive list of items measuring the three domains we examined relevant topic areas from which we could generate newer items. Draft items were identified from highly cited self-report measures of sleep quality, circadian phase, and pathological mood (symptoms of depression and also hypo/mania). Nine sleep scales (e.g., PSQI Buysse et al., 1989 and Insomnia Severity Index Bastien et al., 2001); five circadian rhythm questionnaires [e.g., Munich Chronotype Questionnaire (Roenneberg et al., 2003) and Morningness-Eveningness Questionnaire (Horne and Östberg, 1976)], and 20 mood questionnaires [e.g., BDI-II (Beck et al., 1996), Centre for Epidemiological Studies Depression Scale (Radloff, 1977), and The Hospital Anxiety and Depression Scale (Zigmond and Snaith, 1983)] were used. As part of this generative process we created items that had purified wording, with the specific aim of decreasing overlap between the three domains. Items were selected for being face-valid as relatively pure measures of only one of the three processes. Draft items were then reworded to permit a shared response format (below) and/or to improve clarity and brevity. The preliminary item pool contained 170 items, n = 58 measuring sleep quality, n = 60 measuring circadian phase, and n = 52 measuring problems with mood (full list included in the Supplementary Materials).

The given question prompt was: “The following questions ask about your sleep, mood and timing of daily activities. Pick the answer which best describes you over the past two weeks.” A 2-week timeframe was deemed appropriate, given the nature of the processes under investigation (e.g., 2 weeks is the minimum timeframe required for DSM-5 diagnosis of Major Depressive Episode (American Psychiatric Association, 2013). A 6-point Likert-type response scale (Strongly Disagree to Strongly Agree) was selected for the new questionnaire. We chose not to have a middle response option to avoid ambiguity, as middle response options can reflect socially desirable responding, indifference, uncertainty, or non-applicability (Garland, 1991; Nowlis et al., 2002). Six response categories was considered an optimal balance between the higher validity associated with increasing response options and participant response burden (Chang, 1994; Preston and Colman, 2000).

The expected three factor structure was forced onto the item set. For completeness, a range of empirical methods was also used to explore the number of factors to extract (O'Connor, 2000). The scree plot (Tabachnick and Fidell, 2013), minimum average partial test (Velicer, 1976), parallel analysis (Horn, 1965), and Kaiser-Guttman's criteria (Tabachnick and Fidell, 2013) were all examined. Using an orthogonal rotation (varimax), the adequacy of all factor solutions was compared against the following criteria: item communality magnitudes of over 0.40, cross-loadings less than 0.32 and at least five strongly loading (>0.50) items on each factor (Costello and Osborne, 2005).

To generate the three scale questionnaire, two item reduction steps were applied. First, items were removed if they failed to load with sufficient strength (< 0.32) on any factor, had low communalities (<0.2), or had high cross-loading (>0.3) (Tabachnick and Fidell, 2013). The item content of the final three-scale questionnaire was developed from the remaining 98 items using the following principles: (1) each scale should have the minimum number of items for internal reliability, (2) each scale should contain reverse-coded items, and equal numbers of reverse-coded items were required for each scale, and (3) heterogeneity of meaning across items was sought to cover the breadth of the domain being measured. When items with comparable loadings were semantically similar, brevity and face validity were preferred.

Participants completed standard demographic questions including age and gender, and were asked dichotomous (Yes/No) questions about physical and mental health problems. For example, “Have you ever been diagnosed with a mental disorder”? Affirmative responses were followed by a request for participants to describe the mental disorder in an open response format. Age, gender, physical health problems, history of mental illness, and self-reported difficulties sleeping were investigated as external correlates of the final questionnaire.

A predominantly university student sample (18 years or older) was recruited. A total of 890 individuals commenced the questionnaire package online, with 783 (88%) complete responses recorded (see Table 1 for sample characteristics). Participants were largely first-year university students who were participating in a research experience program (which included on-campus and online students). Other recruitment methods included advertisements on social media and through contacts of the researchers.

A questionnaire package was developed and delivered online using Qualtrics software (https://www.qualtrics.com/au/), Version 15. Participants were given a direct link to the survey. Order of items in the 170-item pool was randomized prior to participant administration. A single item measuring mindless responding was included: “Some participants don't read questionnaires carefully, please select ‘agree’ for this question.” Ninety-two percent of participants selected the correct response for this item. The 62 incorrect responses were excluded leaving a final sample of 720 participants for item reduction analyses [adequate for an EFA of 170 items Tabachnick and Fidell, 2013]. The university's internal ethics review board approved study procedures.

Preliminary analyses showed the data were amenable to factor analysis. Inspection of the correlation matrix at the item generation stage showed numerous correlations >0.3 with sampling adequacy meeting criteria (Kaiser-Meyer-Olkin = 0.94). Bartlett's test of sphericity indicated that the observed correlation matrix deviated significantly from the identity matrix [χ(14, 365)2 = 84692.54, p < 0.001].

The enforced orthogonal three-factor solution (explaining 30% of the total variance) was robust against Costello and Osborne's [35] criteria. Inspection of item content and factor loadings in the forced 3-factor solution generated three robust factors with item content of depressed mood, morningness, and sleep quality.

Data-driven approaches to factor extraction generated a wide range of number of factors to extract. Inspection of the scree plot indicated five factors, the minimum average partial test indicated 26 factors, parallel analysis indicated 26 factors, and Kaiser-Guttman's criteria indicated 35 components with eigenvalues greater than 1. None of the empirically derived solutions were robust against the criteria of Costello and Osborne (2005). The 35-factor solution had five stable factors, the 26-factor solutions had four stable factors and the five-factor solution had three stable factors all explored for item content. All factor solutions indicated factor content of: depressed mood, morningness, and sleep quality; the fourth factor in the 26- and 35-factor solution included item content of value and beliefs about sleep, and the 35-factor model had a fifth stable factor with items pertaining to hypomania.

After initial item removal using the above criteria, 36 items remained on a provisional depressed mood scale, 24 items on a provisional sleep quality scale, and 38 items on a provisional morningness scale. Applying the principles outlined above to select the final items, the scales could be unambiguously named Depressed Mood, Morningness, and Good Sleep on the basis of their item content.

The 15-item Sleep, Circadian Rhythm and Mood (SCRAM) questionnaire had three 5-item scales (one reverse-scored item per scale) (Appendix). As shown in Table 2, these items had moderate-large loadings on their respective factors, and the resultant scales had adequate internal reliabilities, and interpretable intercorrelations in this sample (Table 3).

Figure 1 shows that scores on the SCRAM Morningness and Good Sleep scales were relatively normally distributed, while a non-significant positive skew was observed for scores on the Depressed Mood scale.

Women scored significantly higher on scores of Morningness than men [t₍₇₁₇₎ = 2.77, p = 0.005, Hedge's g = 0.27]; and a non-significant trend was observed for women reporting lower levels of Good Sleep [t₍₇₁₇₎ = 1.78, p = 0.075]. Scores on Depressed Mood did not differ by gender [t₍₇₁₇₎ = 0.053, p = 0.96]. Age was associated positively with Morningness [r₍₇₁₉₎ = 0.21, p < 0.001], and negatively with Depressed Mood [r₍₇₁₉₎ = −0.13, p = 0.001].

Independent samples t-tests were performed to investigate differences in Morningness, Good Sleep, and Depressed Mood in dichotomous self-reported health outcomes for mental illnesses (31% of the sample), reported physical complaints (13%) and sleep problems (36%) (see Table 4).

Self-reporting a diagnosed mental illness was associated with lower scores on Morningness [t₍₇₁₈₎ = 2.55, p = 0.011, Hedge's g = 0.21] and Good Sleep [t_(390.22) = 6.07, p < 0.001, Hedge's g = 0.51], and higher scores on Depressed Mood [t₍₇₁₈₎ = 4.97, p < 0.001, Hedge's g = 0.40]. Self-reported presence of physical complaints was associated with lower scores on Good Sleep [t_(114.65) = 4.23, p < 0.001, Hedge's g = 0.52] and Morningness [t₍₇₁₈₎ = 2.10, p = 0.036, Hedge's g = 0.23], but was unrelated to Depressed Mood [t₍₇₁₈₎ = 1.07, p = 0.28]. Finally, self-reported problems sleeping were associated with lower scores on Good Sleep [t₍₇₁₈₎ = 18.84, p < 0.001, Hedge's g = 1.46] and Morningness [t_(491.89) = 3.57 p < 0.001, Hedge's g = 0.29] and higher scores on Depressed Mood [t_(470.83) = 6.62, p < 0.001, Hedge's g = 0.54].

A predominantly student population was recruited with 498 complete survey responses (95% completion rate). A further 36 participants incorrectly answered the validity question (see Study 1 Method) and were deleted from further analyses. Administration of the questionnaire was identical to Study 1 with participants answering demographic questions and then the draft item pool. From these items the 15 items selected in Study 1 for the SCRAM questionnaire were screened for univariate outliers with 32 participants deleted across the scales to leave a final sample of 430 for the CFA (see Table 1 for sample characteristics).

Using Mplus version 7.11 (Muthén and Muthén, 2013), a CFA with a maximum likelihood estimation was used to investigate the three-factor model found in Study 1. Multiple methods were used to assess model fit (Hu and Bentler, 1999) including: the chi square test of model fit (divided by the degrees of freedom given the sensitivity to sample size), root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), comparative goodness-of-fit index (CFI), and Tucker-Lewis index (TLI). The chi square statistic divided by the degrees of freedom should be less than 3 for an acceptable fit (Schermelleh-Engel et al., 2003), RMSEA scores less than 0.05 reflect a good fit, scores less than 0.08 an adequate model fit; and if the upper limit of the 90% confidence interval is 0.08 this adds additional support of model adequacy (Browne and Cudeck, 1992). SRMR scores should be below 0.08 (Hu and Bentler, 1999; Brown, 2015). CFI and TLI values should preferably be above 0.95, however indices in the range of 0.90–0.95 are acceptable (Hu and Bentler, 1999).

Relationships amongst the three SCRAM scales were found to be congruent with existing literature. First, Study 1 and Study 2 data showed that Depressed Mood was correlated with lower levels of Good Sleep. Sleep disturbances affect mood (Pilcher and Huffcutt, 1996), and mood problems have been associated with lowered sleep quality through increased rumination and subsequent arousal at night (Schmidt et al., 2011). The negative correlation between Morningness and Depressed Mood aligns with data suggesting that evening-type individuals are more likely to experience depression, and report more severe depressive symptomatology (Abe et al., 2011; Chan et al., 2014). Morningness and Good Sleep had a moderate positive correlation, consistent with evidence that eveningness is associated with higher levels of insomnia complaints (Wittmann et al., 2006; Chan et al., 2014; Barclay et al., 2016). Consistent with some previous research (e.g., Roenneberg et al., 2003), Morningness on the SCRAM questionnaire was positively associated with older age and female gender (cf. Merikanto et al., 2012, for evidence that men may have higher levels of morningness relative to women).

In a preliminary assessment of convergent and divergent validity (Study 1), the SCRAM scales showed interpretable relationships to self-reported medical, sleep and psychiatric diagnoses. Previous or current mental illness were associated with lower scores on Morningness and Good Sleep scales, and higher scores on the Depressed Mood scale, consistent with evidence for sleep and circadian rhythm disturbances in a range of mental health disorders, (e.g., Harvey et al., 2011; Gershon et al., 2012, 2015; Pritchett et al., 2012; Soehner et al., 2016; Wong et al., 2016). Self-report of mental illness was associated with higher levels of Depressed Mood; aligning with previous data showing mood complaints to be common in mood and anxiety disorders, and schizophrenia (Levinson et al., 1999; Ohayon and Roth, 2003; Taylor et al., 2005; Pritchett et al., 2012). Finally, consistent with links between physical illnesses and biological rhythm function (e.g., Katon, 2003; Ancoli-Israel et al., 2006; Musiek et al., 2015; Grimaldi et al., 2016; Haus et al., 2016), self-reported physical illness was associated lower scores on the Good Sleep and Morningness scales.

The potential clinical significance of distinguishing between sleep quality, circadian phase, and mood is underscored by the SCRAM scores of participants who self-reported sleeping problems in Study 1. Self-reported sleep problems, not surprisingly, were associated with lower scores on the SCRAM Good Sleep scale but were also associated with higher scores on the SCRAM Depressed Mood scale. This is consistent with sleep disturbance being part of the diagnostic constellation of mood disorders in the DSM-5 (American Psychiatric Association, 2013) and higher negative mood levels observed following sleep disturbance in healthy individuals (Tempesta et al., 2015; Raniti et al., 2016). Self-reported sleep problems were also associated with lower scores on the SCRAM Morningness scale, consistent with an association between morningness and better sleep quality (e.g., Wittmann et al., 2006). The significance of this is that when a patient presents with sleep complaints mood problems and circadian phase alignment are critical in both the remittance of sleep problems and maintenance of sleep quality (Harvey et al., 2005; Carney et al., 2009; Crowe et al., 2016). Sleep quality may be effectively treated but if the patient continues to have a delayed sleep cycle or depressed mood these may serve as independent risk factors for future problems with sleep quality, circadian phase, and/or mood.

With additional validation (see below) the SCRAM questionnaire could improve clinical assessment of symptomatology across the three domains. By distinguishing these three strongly interrelated processes, the SCRAM questionnaire may also be a useful research tool. Malhi and Kuiper (2013) highlight that the substrates of sleep quality, circadian phase, and mood are likely to have different configurations and relationships across psychopathologies. The well-recognized biobehavioral interplay between sleep, circadian function and mood may speak to an under-appreciated superordinate construct or biobehavioral process that could be measured by the SCRAM questionnaire: Murray and colleagues have proposed the Circadian Reward Rhythm as a fundamental motivational process with circadian, sleep and mood manifestations (Murray et al., 2009; see also Alloy et al., 2017).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.