A web-based survey design was used for data collection. The study was reviewed and approved by the Macquarie University Human Research Ethics Committee (MQ HREC: 5201700988), and informed consent was obtained from all participants. All methods were performed in accordance with the relevant guidelines and regulations.

The data sets used and analyzed during this study are available from the corresponding author (NT) upon reasonable request.

An initial list of items was derived from a review of the literature on actions associated with psychological skills identified through multiple sources. These included reviews of (1) skills typically taught in psychological therapies (eg, cognitive skills, behavioral activation, exposure, goal setting, problem solving, social skills training, applied relaxation, mindfulness, acceptance, gratitude, and kindness); (2) actions associated with psychological health, such as healthy daily routines (eg, sleep hygiene, nutrition, and moderated use of electronic devices), physical health (eg, exercise, yoga, stretching, and walking), social activity (eg, talking to friends and spending time with people), meaningful and satisfying activities (eg, doing fun activities), and spiritual and religious activities; (3) symptoms of high prevalence mental disorders as described in psychiatric diagnostic systems [5,6]; (4) previous questionnaires developed by the authors [20], other questionnaires, or similar lists [17-19,22,23]; and (5) consultations with colleagues.

Our research team, consisting of psychiatrists, clinical psychologists, and data analysts with experience in questionnaire development, created a preliminary list comprising >500 actions. Each action was phrased as a verb, and references to psychological techniques or orientations, such as practicing mindfulness, were avoided. During the planning phase, we met regularly to identify, review, edit, and categorize the list, which was reduced to 272 items after duplicates and overlapping items were removed. The items were categorized according to the mechanisms or processes associated with each item; that is, a priori attempts were made to identify theoretical clusters. For example, some items were categorized as primarily concerned with cognitive actions (primary cluster), and within this cluster, they were subdivided into actions associated with challenging unhelpful thoughts, focusing on the future (secondary clusters). In instances where items belonged to multiple clusters, categorization was based on consensus. To capture a broad range of items, the list was shared with colleagues outside the research group, who made further recommendations. The final list of 96 items was based on the following criteria: (1) actions that can be performed daily; (2) measurable, that is, a person can be expected to reliably estimate the frequency of actions over a week; (3) can be completed by most adults; (4) described as a positive action, that is, an action that can be considered a strength; (5) modifiable, that is, can be changed by the person; and (6) did not duplicate another item.

A 5-point Likert rating scale was used to ask participants how often they performed these behaviors over the past week by using the following scoring system: 0=Not at all (0 days per week); 1=1 or 2 days per week; 2=3 to 4 days per week; 3=5 to 6 days per week; and 4=every day (7 days per week). None of the items were reverse scored. After completing the TYDQ, participants were also invited to nominate other actions that were not included in the 96 items they used to improve their mental health.

The survey was promoted to adults across Australia via websites and web-based newsletters from Australian mental health services and Facebook posts and advertisements. Advertisements invited people to participate in a study that aimed to identify activities and habits that affect mental health. Consenting participants provided demographic details and completed web-based questionnaires. Participants were required to be aged ≥18 years. No other exclusion criteria were applied. Details of the sample are included in the Results section.

Three standardized outcome measures were used to evaluate and rank the association of the items with psychological well-being.

The Patient Health Questionnaire 9-item (PHQ-9) scale measures the occurrence of Diagnostic and Statistical Manual IV–congruent depressive symptoms over the past 2 weeks by using a 4-point Likert scale ranging from 0 (not at all) to 3 (nearly every day). Higher scores indicate greater symptom severity [33]. The PHQ-9 has good internal consistency and is sensitive to change [34], with a total score of ≥10 usually but not always associated with a diagnosis of major depressive disorder [35,36]. The Cronbach α in this study was .90.

The Generalized Anxiety Disorder 7-item (GAD-7) scale measures the occurrence of general anxiety symptoms over the past 2 weeks by using a 4-point Likert scale ranging from 0 (not at all) to 3 (nearly every day). The GAD-7 is sensitive to Diagnostic and Statistical Manual IV–congruent GAD, social phobia, and panic disorder, with increasing scores indicating greater severity of symptoms [34,37]. The GAD-7 has been shown to demonstrate sound psychometric properties, with a total score ≥10 usually indicating a likely diagnosis of an anxiety disorder [38]. The Cronbach α in this study was .92.

The Satisfaction With Life Scale (SWLS) is a 5-item scale that measures attitudes toward life satisfaction by using a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree), with higher scores indicating greater life satisfaction. The SWLS has demonstrated good psychometric properties [39,40] and has been used extensively as a measure of life satisfaction in mental health research [41]. The Cronbach α of the SWLS in this study was .90.

The examination of the relationship between the TYDQ items and the standardized scales was operationalized in three detailed and comprehensive steps: (1) item-level analysis, (2) factor analysis, and (3) generalizability of factor solutions.

First, a series of univariate regression models (ANOVA) was used to quantify and statistically test the association between each of the TYDQ item scores and the PHQ-9, GAD-7, and SWLS outcome scores. The strength of the association between items and outcomes was summarized with an R² metric, noting the proportion of the variance explained by each TYDQ item and each outcome. The R² values were used to weigh and rank the 96 items.

To further visualize the relative importance of different TYDQ items, each possible TYDQ item score, ranging from 0 (not at all) to 4 (daily), was tabulated with its corresponding PHQ-9, GAD-7, and SWLS outcomes to form a heat map [42]. Comparing the corresponding outcome scores with the range of TYDQ values enabled the identification of the minimal weekly behavioral frequency score needed to achieve optimal PHQ-9, GAD-7, and SWLS outcome scores for each TYDQ item. The optimal TYDQ score was defined as the minimum weekly behavioral frequency beyond which no additional statistically significant improvement can be observed.

Second, a series of exploratory factor analyses (EFAs) was used to examine how different TYDQ items formed composite latent factors of behavior. An initial EFA was used to identify the factors based on a full list of 96 items. An additional 3 EFAs were conducted to identify latent factors among the subsets of TYDQ items that were associated more closely with the PHQ-9, GAD-7, and SWLS. These EFAs analyzed the subsets of items based on a minimum of 5%, 10%, and 15% R² relationship to any one mental health outcome with the aim of identifying more parsimonious lists of factors closely related to psychological health. A factor was considered viable when the item set comprised a composite eigenvalue greater than 1. Additional diagnostic analyses were used to assess each of the identified factors, including item reliability analysis, mean item intercorrelation, and the ability of the identified factors to replicate each item (ie, commonalities).

All identified factors were tested and ranked for their strength of association with the 3 outcomes by using the same analytics and visualization approach used to evaluate TYDQ items. To further evaluate the importance of different behavioral TYDQ factors for mental health, a series of binary logistic regressions was conducted to test the association between the factor scores and the probability of an individual presenting with PHQ-9 or GAD-7 scores that would be considered within the clinical range.

Each factor was assigned a proposed label, which aimed to represent the overall theme of those items. Consensus regarding the factor label was achieved through discussion and debate, considering other labels could have been used.

A series of confirmatory factor analyses (CFAs) were conducted to examine the reliability of the EFAs within a sample cross-validation analysis (5 randomized subsamples) and the validity of the factor structure along with key subgroups that differed based on age (<30, 30-44, 45-60, >60 years), gender (male, female, and other), the severity of depressive or anxiety symptoms (minimal-mild, moderate, and severe), an indication of employment, and tertiary education. These CFA tests followed the methodology and reporting conventions outlined by Putnick and Bornstein [43] for the examination of measurement invariance, including the examination of configural invariance, metric invariance, scalar invariance, and strict invariance. From the viewpoint of scale development, this step sought to evaluate the reliability and generalizability of the identified factor solutions along relevant subgroups in clinical research (dimensionality).

All analyses were performed using R statistical software version 4.1.1 [44] and the Lavaan package [45]. A conservative P value of .01 was considered the threshold for statistical significance, reflecting the large sample and the multiple number of tests conducted. In all analyses, participants with missing data were not included in the analyses owing to difficulty imputing unbiased missing outcomes in a high-dimensional analysis and cross-sectional settings. All EFAs and CFAs were based on weighted least-squares estimators to account for the ordinal categorical TYDQ item score range. For all EFAs, with the aim of identifying unique subsets of items, a varimax factor rotation was adopted to maximize the number of factors identified and minimize item-factor cross-loadings [46].

Items with factor loadings <0.50, a threshold that aimed to achieve a conservative balance between recommendations in the literature [32,47], were suppressed as were items with suboptimal association with their corresponding factor (accounting for <25% of the factor variance) [48].

Participants in study 1 were recruited from March to June 2018, during which time 3755 people consented to participate and 3040 (80.96%) completed the questionnaire. Sample characteristics are presented in Table 1. Participants had a broad range of both demographics (age, gender, location, education, and employment) and mental health symptoms. Approximately half of the sample scored above the clinical cutoffs on the PHQ-9 (n=1710, 48.1%), GAD-7 (n=2020, 57.9%), and SWLS (n=2188, 64%), allowing the testing of the association of everyday behaviors for those above and below recognized clinical cutoffs on the PHQ-9 and GAD-7.

Multimedia Appendix 1 includes each of the 96 items ordered based on the a priori primary and secondary theoretical clusters created during item development and based on the observed average weekly score. Multimedia Appendix 1 also uses a heatmap to indicate the frequency with which participants reported performing each item each week. For example, the first item in Multimedia Appendix 1 was the 13th item presented in the survey and read I read, listened, or watched something I enjoyed. A priori, this item was categorized as an Activity/Meaning (primary and secondary clusters), and the average weekly score was 2.72 (mean TYDQ weekly score), indicating that the mean frequency reported was between 4 and 6 days each week. The final column indicates that 37.4% (n=1404) of the sample reported doing this action daily, whereas 3.7% (n=139) reported not doing this action in the previous 7 days.

As shown in Multimedia Appendix 1, the frequency of actions (TYDQ mean scores) ranged across and within clusters. For example, 2 of the highest frequency scores were observed within the Healthy Routine cluster, with daily showers or baths reported by 69.3% (n=2602) of the sample, and avoidance of illicit drugs or misuse of medication reported by 83.3% (n=3128).

The 96 items were then subjected to tests of the strength of association with each of the PHQ-9, GAD-7, and SWLS outcomes. The results from this first series of analyses are presented in Multimedia Appendix 2 for each item, with items grouped around the a priori clusters. The columns of the table report the mean PHQ-9, GAD-7, and SWLS scores observed with differing rates of weekly behavioral frequency. The PHQ-9, GAD-7, and SWLS scores were further incorporated into a heat map visualization to highlight the TYDQ items associated with the highest symptom scores among the 96 TYDQ items, where red hues indicate deterioration of psychological health (ie, symptom scores and satisfaction with life) and blue hues indicate improved psychological health.

The proportion of symptom variance explained (R²) was reported as a summary measure of the association between each item and outcome. In Multimedia Appendix 2, the presentation order of the 96 items illustrates the relative ability of TYDQ items to account for the cumulative variance of the 3 outcomes, with the strongest items reported at the top of each a priori cluster. Together, the 96 items were significantly associated with each of the 3 outcomes (TYD outcome regression β test resulting in P<.001), except item TYDQ 57 (I fulfilled my responsibilities even though I didn’t want to) and its association with GAD-7 outcomes. Most items showed a linear relationship between higher frequency and lower symptom scores, with a minority of items showing a flattened, convexed, or positive relationship with PHQ-9 and GAD-7 outcomes. However, the strength of the association between each of the items and each outcome varied substantially, both within and across clusters. For example, TYDQ 16, I treated myself with respect, illustrated the strongest association with depressive symptoms (accounting for 22.5% of all PHQ-9 scores); TYDQ 70, I kept a realistic perspective on things, was the strongest correlate of anxiety symptoms (accounting for 21.1% of all GAD-7 scores); and TYDQ 54, I had something to look forward to, was the strongest correlate of satisfaction with life (accounting for 25.9% of all SWLS scores). Across the items, the threshold of activity associated with optimal well-being varied from half a week to daily, with the majority of items requiring frequent (almost every day) but not necessarily daily activity.

EFAs were conducted to identify underlying groupings among the items and to explore a brief, latent factor structure of behaviors associated with psychological health. The results from the EFA of the complete 96-item list are presented in Multimedia Appendix 3. In total, 16 factors were identified, with 56 of the items demonstrating factor loadings above 0.5 for any 1 factor. Factors with loadings of ≤0.5 were suppressed. Across the TYDQ list, the factors appeared to form along their a priori clusters, with most items grouped within their clusters. No item-factor cross-loadings were identified above the loading of 0.5.

The results of the tests of association between the resulting factor scores and each of the PHQ-9, GAD-7, and SWLS outcomes are further reported in Multimedia Appendix 3, which describes the outcomes of a series of 4 EFAs, each testing a different number of items (see Multimedia Appendix 3 for items). This shows that the complete list of 16 factors, based on 56 items, accounted for 41%, 37%, and 32% of the outcome R², respectively. The resulting 16 factors can also be seen as varying in their strength of association with the 3 outcomes, with estimates of R² ranging from <0.1% to 24%.

The EFA was repeated with another subset of items (68 items), selected based on the individual item’s ability to account for at least 5% of any of the 3 symptom outcomes (R²>5%), which provided an opportunity to examine patterns among TYDQ items more closely related to psychological health. The analysis identified 11 factors, with only 45 items demonstrating factor loadings >0.5 for any one factor. The items were clustered within their respective a priori clusters, and no item-factor cross-loadings were identified. A test of the association between the resulting factor scores and each of the PHQ-9, GAD-7, and SWLS outcomes illustrated that a complete list of 11 factors based on 45 items accounted for 33%, 29%, and 32% of the outcome R², respectively, with each of the individual factors varying in their strength of association from 5% to 21% (Multimedia Appendix 4).

In a third EFA, an even briefer subset of items (35 items) was selected based on the item’s ability to account for a minimum of 10% of any of the 3 symptom outcomes. This resulted in 5 factors and 27 items with a factor loading above 0.5 for any 1 factor. No item-factor cross-loadings were identified. A test of association between the resulting factor scores and each of the PHQ-9, GAD-7, and SWLS outcomes revealed a list of 5 factors based on 27 items, which accounted for 38%, 31%, and 33% of the outcomes R², respectively (Multimedia Appendix 4). The formation of factors largely followed the a priori clusters, and the resulting factors varied 3-fold in their strength of association with each of the outcomes (9%-28%).

A fourth EFA, which included items, identified a stringent R²>15% criterion. This resulted in a brief set of items (8 items) found to represent a single composite factor combining key items from several domains. Although this result is interesting, the PHQ-9, GAD-7, and SWLS variance explained by a single factor (Multimedia Appendix 4) was suboptimal (R²=36.6%, R²=26.7%, and R²=32.9%, respectively), and the representation of multiple domains, including activity planning, self-representation, and values through a single nonspecific factor, would inevitably limit the ability to assess and interpret the type of behavior.

Together, these EFAs illustrate that the 96-item list could be reduced to a more parsimonious and selective item list, with an optimal 5-factor, 27-item solution that retained >90% of the strength of association with each of the symptom outcomes, with only a quarter of the items (27/96, 28%). Each factor solution was further assessed for item reliability, mean item intercorrelation within each factor, and the ability of the identified factors to replicate each item (commonalities), as collated in Multimedia Appendix 5.

A series of CFAs were conducted to examine the measurement invariance of the 5-factor solution as a brief but sensitive measure of psychological well-being across 5 categories, namely age groups, gender, depression and anxiety symptoms, education, and employment. The results from the CFAs are collated in Multimedia Appendix 6. The identification of TYDQ differences in the intercept and residual scores indicated that gender, age, and baseline symptoms were associated with different frequencies of weekly activity, although the TYDQ scores formed the same latent patterns across these groups. To check for the potential of alternate models, we reran the EFA within each of the age, sex, and PHQ-9 subgroups. The results identified the 5-factor solution reliably replicated as the most prominent latent solution when no parameter restraints were made or when the rotation methodology was changed (Multimedia Appendix 7). Together, the results from these CFAs illustrate that the latent 5-factor structure identified was statistically reliable, generalized across subgroups of clinical interest, and replicated across differing aspects of the statistical methodology (rotation and parameter constraint).

The design of study 2 replicated that used in study 1. Ethics approval for data collection was obtained from the Macquarie University Human Research Ethics Committee (MQ HREC: 5201700988), and informed consent was obtained from all participants.

Study 2 evaluated a version of the TYDQ comprising 59 of the 96 original items (Multimedia Appendix 8), with each item using the same 5-point rating scale used in study 1. This brief list was constructed in several steps. First, items that accounted for at least 5% of the PHQ-9, GAD-7, and SWLS were identified, resulting in an initial list of 67 items. Second, this list was edited to remove items with similar wording or duplicated items, resulting in the removal of 23 items. Third, 15 additional items from the original 96 item list were added to test the association between items, such as kindness to others and gratitude, which were considered clinically important and expected in study 1 to be associated with the outcomes but were not. We reviewed each of these steps.

The procedure used in study 1 was repeated in study 2. As shown in Table 1, a total of 3756 participants consented to participate in the study and started filling the questionnaires, which were available between June and August 2020. These dates coincided with social and travel restrictions across Australia owing to the impact of the COVID-19 pandemic.

The outcome measures used in study 1 were also used in study 2.

The three analytical steps in study 1 were repeated: (1) item-level analyses, (2) factor formation and analyses, and (3) analyses that explored the generalizability of factor solutions across different subgroups. Additional analyses were conducted to compare study 1 and study 2 samples on TYDQ scores, the strength of association between the items and symptom outcomes, and factors. For example, the comparison of TYDQ scores from the 2 studies was operationalized by treating the studies as a group variable and regressing this variable on the TYDQ item scores. Similarly, the association between TYDQ items and symptom outcomes was tested by regressing the TYDQ item score, group, and item-by-group interaction on outcome measures.

A total of 3756 people consented to participate in study 2, and 3160 (84.13%) completed all the questionnaires. Participant characteristics and comparison with the study 1 sample are presented in Multimedia Appendix 8. These results indicate significant differences between the samples across most demographic variables, symptom severity, satisfaction with life, use of medications for anxiety or depression, and mental health service use.

The TYDQ score for each item, the association of each TYDQ item with the 3 symptom outcomes, and testing of differences between the 2 study samples are included in Multimedia Appendix 8. The results indicated a significant increase in the average score of most TYDQ items from study 1 to study 2. The results also indicate that the strength of association between most TYDQ items and each of the 3 outcomes was greater in study 2 than in study 1. However, the relative strength of association across and between the items and the outcomes was similar in magnitude across the samples, as seen in the heat map.

In the second step, the 59 items were subjected to a dimension-reduction analysis. The complete list of 59 items identified a 10-factor solution when no parameter constraints were imposed on the data, with 9 of these factors overlapping with the study 1 EFA solution (Multimedia Appendix 9). Using the inclusion criterion of R²>10%, a 26-item solution was identified using the data from study 2, replicating the pattern of the 5 factors identified in study 1. Using the inclusion criteria of R²>10% and reanalyzing the results of study 1 resulted in an even more parsimonious 21-item solution (Multimedia Appendix 9) that reliably retained the factor structure. The 5 factors and the 21-item solution identified in the reanalysis of study 1 results (Multimedia Appendix 9; inclusion criteria of R²>10%) were then compared with the same 21 items from the study 2 results, and associations between the 5 factors and the 3 outcome variables were compared across samples (Multimedia Appendix 10). This demonstrated close similarities in strength, directionality, significance, and optimal frequency cutoffs between the 2 samples.

A set of 3 EFAs were then conducted for item subsets with no selection criteria, inclusion criteria of R²>10%, and inclusion criteria of R²>15%. The 5-factor 21-item solution associated with the R²>10% criteria in study 2 accounted for the highest amount of PHQ-9, GAD-7, and SWLS outcome variance (R²=46.8%, R²=38.2%, and R²=38.1%, respectively). Consistent with this, a reanalysis of the data from study 1 using the same R²>10% criteria demonstrated that this 5-factor 21-item solution accounted for the highest amount of PHQ-9, GAD-7, and SWLS outcome variance (R²=37.1%, R²=30.0%, and R²=30.5%, respectively).

CFAs seeking to verify a 5-factor 21-item solution across the study 1 and study 2 samples resulted in metric invariance (similarities in factors and item loading) but not scalar or strict invariance (Multimedia Appendix 11). This result is consistent with the interpretation that participants from the 2 studies differed in their means but not item or factor importance.

In brief, the overall results of study 2 replicated the results and 5-factor model obtained in study 1, albeit with a 21- rather than a 27-item solution. These results confirm the strength of the relationship between key items in the TYDQ and psychological health outcomes and the underlying factor structure.