The Hazards of Daily Stressors: Comparing the Experiences of Sexual and Gender Minority Young Adults to Cisgender Heterosexual Young Adults During the COVID-19 Pandemic

The most common research method for examining stressor exposure in daily life is via ambulatory assessment methods such as experience sampling or daily diaries that involve participants responding to brief surveys via mobile phone as they go about their daily lives (Almeida, 2005). The many benefits of this method include reducing recall bias due to measurement in close temporal proximity to events in daily life and high ecological validity as the measurements are taken in situ. However, event-based data (i.e., whether and when specific types of stressor events occurred) collected in this way have a feature called right censoring that, ideally, will be factored into the analytic approach used when event-based variables are used as dependent variables in statistical analyses (Mills, 2011). Right censoring refers to participants who did not experience the event of interest (e.g., a specific type of stressor) during the period they participated in the study. When applying statistical approaches that do not account for right censoring (e.g., t-tests, linear regressions, and other approaches within the general linear model) to right-censored dependent variables—which is common in this area of research when testing group differences on event frequency (Almeida & Horn, 2004; Wardecker et al., 2022)—researchers may unwittingly make an important assumption that right-censored cases never experienced the event, even after the observation period ended. We argue that this assumption is untenable in the realm of frequent and common daily stressors such as conflicts, and that we therefore need to employ statistical approaches that account for right censoring when we are interested in predicting the risk of stressor exposure.

Survival analysis is a form of regression that accounts for right-censored dependent variables (Lougheed et al., 2019; Mills, 2011; Singer & Willett, 2003). It is a method for estimating the likelihood and occurrence of events over time. This approach originates in fields of public health, epidemiology, and medicine where researchers are often interested in predicting the likelihood of medical events (e.g., infection, death), hence the parlance of predicting who “survives” (Kleinbaum & Klein, 2012; Mills, 2011). Survival analysis can be used to estimate the timing of single events with only one occurrence (e.g., death), and it can also be used in a multilevel framework to estimate events that can recur over time (e.g., stressors common in daily life; Kleinbaum & Klein, 2012; Lougheed et al., 2019; Stoolmiller & Snyder, 2006).

Right-censored cases in the survival analysis framework contribute to the “risk set”—participants who have yet to experience the event and are therefore at risk for it—in the calculation of the risk of event occurrence at a given time point. In contrast, the more common approaches of calculating mean number of events for use as a dependent variable in a linear regression or t-test (e.g., Wardecker et al., 2022) involves the implicit assumption that right-censored cases (who contributed values of 0 to the dependent variable) never experienced the event, at all. Researchers using these approaches with these types of data need to restrict interpretation to the limited sampling period, which may not be typical or representative for participants; make an unrealistic assumption that participants who did not experience the event of interest during the time they were enrolled in the study never experienced the event even after their participation in the study ended; or accept inaccurate estimations of differences between groups. In the current study, we used multilevel survival analysis (MSA) to compare the risk of exposure to different types of common, recurring stressors in daily life between SGM and CH undergraduate students during the COVID-19 pandemic.

The primary aim of this study was to examine the daily stressors experienced by undergraduate students during the COVID-19 pandemic, the association between the experience of daily stress and same-day negative mood (stressor reactivity), and whether these vary between undergraduate students with SGM identities and their CH counterparts. We explored these topics with two sets of preregistered research questions (see pre-registration, data analysis files and scripts here: https://osf.io/p3fgd:

1. Do the risks of experiencing specific types of daily stressors differ between SGM individuals and CH individuals during the COVID-19 pandemic? We examined group differences on the likelihood of experiencing the following types of stressors: Argument, conflict, or disagreement; family or home stress; work or school stress; financial problem; traffic or transportation stress; health problem or accident; stressful event that happened to close friends or family; other stressors; and any stressor. We hypothesized that SGM individuals would be at greater risk for experiencing daily stressor types of argument, conflict, disagreement; family or home stress; and work or school stress than CH individuals given that the COVID-19 pandemic has created additional interpersonal and school stressors for some SGM adults, such as potential returns to unaccepting home environments with the closure of universities and programs (Gonzales et al., 2020; Newcomb et al., 2019). We did not have specific hypotheses regarding group differences on the risk of experiencing financial problems, traffic or transportation stress, health problems or accidents, stressful events happening to close friends or family, or “other” stressors, so we examined these group differences in an exploratory way.

2. Do SGM individuals experience greater reactivity to daily stressors in terms of same-day reports of negative mood compared to CH individuals? In line with previous research (Wardecker et al., 2022), we hypothesized that SGM individuals would show greater stressor reactivity than CH individuals (i.e., SGM individuals will report greater negative affect than CH individuals on days stressors occur).

It is important to be self-reflexive in research and positionality statements can aid in the interpretation of study results (Roberts et al., 2020). The lead author of this study is a bisexual cisgender Canadian woman who lived for several years in the United States. Her experiences related to sexual orientation, citizenship, and country of residence have afforded her insights that influenced the conceptualization, design, and interpretation of this work.

Participants were 930 undergraduate students recruited from the Department of Psychology participant research pool at a university in western Canada between February, 2021 and March, 2022. Participant recruitment occurred over four semesters. Data from Cohort 1 were collected between February and March 2021, when all courses at the university were held online and the province had strict lockdown guidelines regarding in-person social contact (including limitations to the number of individuals from different households permitted to socialize in person). Data from Cohort 2 were collected between May and June 2021, shortly after the provincial government had announced a loosening of lockdown restrictions and the university had announced a phased approach for returning to campus. Data from Cohort 3 were collected between October and November 2021 after the province had achieved a high rate of uptake of the COVID-19 vaccine and many but not all courses on campus had returned to in-person settings. Data from Cohort 4 were collected between February and March 2022, shortly after the university returned to in-person courses after it had temporarily moved all courses online for the first few weeks of this term as a result of higher rates of Omicron-related COVID-19 infections. Figure 1 summarizes the pandemic-related restrictions at the university and province by cohort.

This study consisted of two parts. In Part 1, participants filled out a one-time series of questionnaires via Qualtrics regarding demographics, mental health symptoms, emotional tendencies, and well-being (see Measures for description of measures used in the current study). Participants were automatically assigned 0.5 course credit via the Department of Psychology Sona Systems site (a research and participant management site) upon the completion of the Part 1 survey, and were emailed details on how to access Part 2 of the survey.

In Part 2, participants completed a daily diary survey via a free mobile app called ExpiWell (Tay & Pineda, 2020) once per day for 14 days. This daily survey consisted of brief questions about daily experiences (exposure to stressors and positive events, daily mood). Participants received a notification to their mobile phone via ExpiWell each day at 8:00 pm, and participants received a reminder 20 minutes later if they had not yet completed the daily report. Participants could submit responses each evening up until 11:59 pm. Specific questionnaires to be completed in Part 2 were the Your Feelings questionnaire to assess daily mood (adapted from Charles et al., 2019) and the Today’s Events questionnaire to assess the experience of specific stressor and positive events (adapted from Sin & Almeida, 2018). Participants were granted credit upon completion of the 14 days, receiving 0.5 credit for completing 7 or fewer daily diaries, and 1.0 credit for completing 8 or more daily diaries.

A total of 930 individuals participated in this study. Of these 930 participants, 49 did not provide the needed ID variable into Part 2 of the study and were therefore excluded because we were not able to match their demographic variables to their daily diary data. Of these 881 participants, 265 completed Part 1 only and did not provide data for the Part 2 daily diaries. An additional 7 participants were excluded due to missing data on key demographic variables. The final sample used in the current study consisted of 609 participants. See Table 1 for a full breakdown of key demographic variables of this sample. Participants’ ages ranged from 17 to 30+ years (mode = 20 years old).

Participants reported on several demographic variables. Table 1 shows the demographic characteristics of the full sample and by cohort. Variables of SGM status, gender, international student status, and ethnicity were examined in the current study. Sexual orientation was measured with the question, “What is your sexual orientation?” with response options asexual, bisexual, gay, lesbian, pansexual, straight/heterosexual, Two Spirit, and An identity not listed (with an option to specify). Gender was measured with the question, “What is your gender?” with response options Gender non-conforming (Genderfluid), Man, Woman, Trans man, Trans woman, Genderqueer, Prefer to self-describe (with an option to specify). We combined individuals who identified as asexual, bisexual, gay, lesbian, pansexual, Two Spirit, and an identity not listed on the sexual orientation question; and individuals who identified as gender non-conforming, trans man, trans woman, genderqueer, and an identity not listed on the gender question into one sexual and gender minority category. Sexual and gender minority identity was coded as 1 and CH individuals were coded as 0. Gender was coded as Male as 1, and Female and other genders as 0.

Ethnicity was measured with the question, “Which of the following responses best identify your ethnic background? (Please select the best description of your background).” Response options included: White, Black, Indigenous, Asian, East Asian, Hispanic, and Multiple ethnicities. Participants who endorsed Indigenous received the following follow-up options: Métis, Inuit, First Nations, and Other (with an option to specify). Individuals who identified as White on the ethnicity question were coded as 0, and all other participants were coded as 1.

Student international status was measured with the question, “Are you an international or domestic student?” with response options International and Domestic. Individuals who indicated that they are international students were coded as 1 and individuals who indicated that they are domestic students were coded as 0.

Cohort was indicated with a series of dummy variables with Cohort 1 as the reference group. The dummy variables were coded as Cohort 2: Cohort 2 = 1, else = 0; Cohort 3 = 1, else = 0; and Cohort 4 = 1, else = 0.

On each day of Part 2 of the study, participants responded to a question pertaining to the types of stressful events they experienced. This questionnaire was based on the Today’s Events questionnaire based on the National Study of Daily Experiences (Sin & Almeida, 2018), the Work, Family, and Health Network Study (Sin et al., 2017), and the ESCAPE Study (Scott et al., 2015). Participants were asked, “Did any of these stressful events occur today? A stressful event is any event, even a minor one, which negatively affected you. (If more than one happened, please select the categories for each event).” Response options were: Argument, conflict, or disagreement; Family or home stress; Work or school stress; Financial problem; Traffic or transportation stress; Health problem or accident; Stressful event that happened to close friends or family; Other stressful event (please describe); No stressful event happened today.

For use as dependent variables in the multilevel survival analysis models, each stressor type (argument, conflict, or disagreement; family or home stress; work or school stress; financial problem; traffic or transportation stress; health problem or accident; stressful event that happened to a family member or friend; other) was coded as 1 on days each participant indicated the event occurred and coded as 0 on days each participants indicated the event did not occur. An “any stressor” variable was calculated to indicate days on which any of the above listed stressors happened, which was coded as 1 on days any stressor occurred, and 0 on days that no stressor occurred. The same “any stressor” variable defined above was used as a level 1 predictor variable in the multilevel model testing research question 2.

Each day of Part 2, participants responded to the Your Feelings questionnaire which contains 16 emotion-word items for participants to rate their daily mood on. This measure was adapted from the National Study of Daily Experience (Charles et al., 2019). Participants rated their responses to the question, “Below is a list of words describing different feelings. How well does each word describe how you felt today?” on a scale from 0 to 4 (0 = None of the time, 1 = A little of the time, 2 = Some of the time, 3 = Most of the time, 4 = All of the time): anxious, sad, angry, frustrated, enthusiastic, happy, disgusted, satisfied, confident, calm, like you belong, close to others, lonely, ashamed, proud, and full of life. Items indicating negative mood were: Anxious, sad, angry, frustrated, disgusted, lonely, and ashamed. The mean of each participants’ reports on all negative mood items (anxious, sad, angry, frustrated, disgusted, lonely, ashamed) will be calculated for each day. Both the within-person and between-person reliability of the daily negative mood were calculated based on the guidelines by Bolger and Laurenceau (2013). These scores show whether there are reliable within-person differences in change over time, and whether there are reliable between-person differences in the measurement of daily negative mood, respectively. The interpretation of the within- and between-person reliability scores are similar to the interpretation of Cronbach’s alpha. The negative mood subscale showed good within-person (ranging from .73 to .77) and between-person (ranging from .74 to .83) reliability across all cohorts.

Hypotheses associated with Research Question 1 were tested with separate multilevel survival analysis (MSA) models using the Cox semiparametric approach (Cox, 1972). MSA is appropriate for use with data that contain right-censored cases (individuals who did not experience the event of interest during the observation period). We ran models only for dependent variables that showed no more than 50% of cases with right censored data in line with recommendations to ensure that the extent of right censoring does not lead to untrustworthy model results (Griffin, 1993) and our preregistered data analysis plan.

We controlled for several variables in each MSA model: gender, domestic versus international student status, ethnicity, and cohort. Equation 1 shows the general form for each of the MSA models we ran:

Equation 1:

In Equation 1, the hazard of the jth episode of a stressor at time interval t in individual i is the product of the baseline hazard h₀(t), an exponentiated random effect v_i for individuals, and an exponentiated linear function of time-invariant predictors of SGM identity (β₁), gender (male versus other, β₂), international student status (β₃), ethnicity (β₄), and dummy-coded comparisons of each subsequent cohort to cohort 1 (Cohort 2 versus Cohort 1 [β₅]; Cohort 3 versus Cohort 1 [β₆]; and Cohort 4 versus Cohort 1 [β₇]).

We fit all MSA models using the coxph() function of the survival package (Therneau, 2015a) in R. For all models, we first fit a baseline model of the hazard function (no independent variables included) so we could examine the shape of the baseline survival function (Lougheed et al., 2019). Then, we added all predictors into the next model and evaluated the goodness of model fit using the likelihood ratio test to determine if the model including all predictors fits significantly better than the baseline model (Mills, 2011).

Hypotheses associated with Research Question 2 were tested with a multilevel model predicting mean daily negative mood, see Equation 2 below.

Equation 2:

The mixed model Equation 2 shows predictors of daily negative mood for time t and individual i including daily stress exposure (within-person, $γ10$⁠, coded as 0 = no stressor day, 1 = stressor day), Time (⁠$γ30$⁠, centered to have a zero mid-point and 1 unit will correspond to 1 week), and weekday versus weekend $γ40$ , weekday = 0 and weekend = 1), SGM identity (γ₀₁, coded as SGM = 1, other = 0), between-person daily stress exposure (γ₀₂, the mean number of days each participant reported at least one stressor), domestic versus international student status (γ₀₆, 0 = domestic, 1 = international), ethnicity (γ₀₅, 0 = White, 1 = non-White), gender (γ₀₄, Male = 1, else = 0), and cohort (in a series of dummy variables with Cohort 1 as the reference group coded as Cohort 2: Cohort 2 = 1, else = 0 [γ₀₇]; Cohort 3 = 1, else = 0 [γ₀₈]; and Cohort 4 = 1, else = 0 [γ₀₉]). We also included a cross-level interaction in which SGM status predicted the slope of the effect from stress to mood (γ₂₀).

First, we examined descriptive characteristics of time until stressor event variables to determine the appropriateness of the data for MSA. Then, we proceeded to build MSA models to test our research questions for variables that met our preregistered criteria of having no more than 50% of right-censored cases in the full sample.

Table 2 shows descriptive statistics for each daily stressor event we measured, including the percentage of right censored cases in the full sample. Given that data with more than 50% right censored cases can be problematic for estimating MSA models (Griffin, 1993), we did not proceed to the next step of analysis for the variables financial problem, traffic or transportation stress, health problem or accident, and stressful event that happened to a family member or friend. Figure 2 shows the survival times separately by SGM and CH individuals for recurring events of argument, conflict, or disagreement; family or home stress; work or school stress; and any stress. Each participant is represented by a horizontal line within groups of stressor episode. The number of survival times (horizontal lines) per individual is equal to their number of stressor episodes. Looking vertically up the graphs, the time until Conflict (Figure 2, Panel A) and Home/Family (Figure 2, Panel B) stressor events appears to decrease as the number of episodes increases, with fewer participants represented in later episodes. For Work/School stressor (Figure 2, Panel C) and Any Stressor events (Figure 2, Panel D), the steep steps on these plots indicate that many participants experienced these stressors early in the study period and many participants experienced these event categories multiple times.

Our final data analysis approach differed from our preregistered plan in two ways. First, we were not able to code gender as a series of dummy variables comparing men and “other” (non-male or female) gender categories with female as the reference group given the overlap with the SGM variable. Thus, we coded gender as a binary variable comparing males to all other genders, given that men tend to be less at risk than all other genders for mental health and stressor-related outcomes (Hoyt et al., 2021). Second, we did not impute missing data via the “last observation carried forward” method (Tonini et al., 2014), which would involve replacing missing data on stressor event variables with the same value of the preceding non-missing value. Given extensive criticism of this approach (e.g., Lachin, 2016), we instead handled missing stressor event data as by using days with missing data to factor in to the calculation of the time until the next reported event but without assuming events on days with missing data. This approach would be expected to result in a more conservative test of event timing (i.e., it would not potentially artificially inflate the significance of event occurrence).

We tested group differences on the risk of experiencing types of stressors in separate models for each of the following dependent variables: (i) argument, conflict, or disagreement; (ii) family or home stress; (iii) work or school stress; and (iv) any stressful event. For all models, we first fit an unconditional model (no predictors included) to obtain the baseline survival function (see Figure 2). Then, we tested our hypotheses by adding predictors and covariates to the model, and examined model fit and diagnostics prior to interpreting model results (Lougheed et al., 2019). We fit the models using the coxph() function of the survival package (Therneau, 2015a) in R, specifying a random effect for individuals with a frailty term and using the Efron approximation of the partial likelihood for estimation. Next, we examined the proportional hazards assumption statistically with the cox.zph function which creates an interaction between each predictor variable and log-transformed time (Therneau, 2015a), with significant deviations of the observed values from the expected values of the predictors indicating nonproportional hazards (Mills, 2011). We examined the goodness of fit for each model using a likelihood ratio test that compares the fit of nested models (unconditional model to model including predictors and covariates; Mills, 2011). This test is a joint test of the random effect variance and the fixed effects of predictors and covariates in our models. Finally, we interpreted the hazard ratio of our model results. Hazard ratios can be interpreted as the magnitude of difference in the risk of event occurrence between two groups being compared. Similarly to odds ratios, values of 1 indicate no association between the predictor and the outcome, values < 1 indicate a lower hazard of event occurrence for higher values of the predictor, and values > 1 indicate a higher hazard of event occurrence at higher values of the predictor (Mills, 2011).

Model 1:Argument, Conflict, or Disagreement. In Model 1, we tested whether the likelihood of experiencing arguments, conflicts, or disagreements differed between SGM and CH individuals (see Table 3). The significant likelihood ratio test indicated that this model including predictors fit significantly better than the unconditional model, X² (188.40, N = 1037 events) = 409.20, p < 0.001. None of the predictor variables violated the proportional hazards assumption (ps ranged from .11 to .71). Contrary to expectations, groups did not differ on the likelihood of experiencing arguments, conflicts, or disagreements (β₁, HR = 1.03). The likelihood of experiencing arguments, conflicts, or disagreements also did not differ by gender, student status, ethnicity, or cohort. Random effects can be interpreted by exponentiating their standard deviation, which indicates the relative risk of an individual who is more (or less) at risk than the prototypical individual (i.e., at 1 SD above the sample mean; Therneau, 2015b). The variance of the random effect was 0.28, and taking the square root of this value gives the standard deviation (SD = 0.53). This value indicates that an individual at high risk for experiencing this stressor (+ 1 SD above the sample mean on risk of an argument, conflict, or disagreement) had 1.70 = exp(0.53) times the risk of experiencing this type of event than did the average individual.

Model 2: Family or Home Stress. In Model 2, we tested whether the likelihood of experiencing family or home stress differed between SGM and CH individuals. The results of Model 2 are shown in Table 3. The significant likelihood ratio test indicated that this model including predictors fit significantly better than the unconditional model, X² (270.10, N = 1222 events) = 668.50, p < 0.001. None of the predictor variables violated the proportional hazards assumption (ps ranged from .18 to .74). Contrary to expectations, groups did not differ on the likelihood of experiencing family or home stress (β₁, HR = 1.13). The likelihood of experiencing family or home stress differed significantly by gender, with males being significantly less likely to report experiencing this type of stressor compared to other genders (β₂, HR = 0.77). Interpreted as a percent change in the hazard, males were 100 x [0.77 -1.00] = -23% less likely to report experiencing family or home stress compared to other genders. The likelihood of experiencing family or home stress did not differ by student status, ethnicity, or cohort. The variance of the random effect (see Table 3) was 0.45, and taking the square root of this value gives the standard deviation (SD = 0.67). This value indicates that an individual at high risk for experiencing this stressor (+ 1 SD above the sample mean on risk of family or home stress) had 1.95 = exp(0.67) times the risk of experiencing this type of event than did the average individual.

Model 3: Work or School Stress. In Model 3, we tested whether the likelihood of experiencing work or school stress differed between SGM and CH individuals. The results of Model 3 are shown in Table 3. The significant likelihood ratio test indicated that this model including predictors did not fit significantly better than the unconditional model, X² (7, N = 4,150 events) = 8.54, p = 0.30, so we did not interpret the results from this model.

Model 4: Any Stressor. In Model 4, we tested whether the likelihood of experiencing any stressor differed between SGM and CH. The results of Model 4 are shown in Table 3. The non-significant likelihood ratio test indicated that this model including predictors did not fit significantly better than the unconditional model, X² (7, N = 5,467 events) = 3.73, p = 0.80, so we did not interpret the results from this model.

Our final model differed from our preregistration in two ways. First, similarly to our MSA models, gender was coded differently as described above. Second, although we described a multilevel equation with two levels in the preregistration, we used a mixed model equation in this manuscript for improved readability.

We used a multilevel modelling approach following the guidelines by Bolger and Laurenceau (2013) to analyze our data in which we examined differences in stressor reactivity between SGM and CH individuals. As Table 4 shows, the effects of between-person and within-person stress were significant on negative mood. Participants who experienced more stressors also experienced greater negative mood (between-person effect, $γ02)$⁠, and the experience of stressors on a given day was associated with experiencing greater-than-average levels of negative mood that same day (within-person effect, $γ10)$⁠. The effect of SGM status on negative mood was also significant (⁠$γ01$⁠), showing that SGM individuals were more likely to experience negative mood in their daily life than CH individuals. However, our hypotheses regarding group differences in stressor reactivity were not supported: Neither the interaction between between-person stress (⁠$γ03$⁠) and SGM status nor within-person stress and SGM status (⁠$γ20$⁠) in predicting negative mood were significant. These results indicate that our sample showed stressor reactivity in general—days with stressors were associated with greater negative mood that same day—although stressor reactivity did not differ between SGM individuals and CH individuals. Finally, the variance of the random of negative mood and the random slope of the effect from within-person stress to negative mood (Cov(⁠$μ0i,μ1i$⁠)) was positive and significant, which indicated that when participants showed higher negative mood, they showed higher stressor reactivity in their daily life.

We examined whether SGM and CH undergraduate students experienced differences in their risk of encountering stressful events in their daily lives, and whether they differed in their reactivity (negative mood) to daily stressors during the COVID-19 pandemic. A growing literature documents the experiences that are associated with mental health disparities experienced by SGM individuals Hoyt et al., 2021; Wardecker et al., 2022; Woodford et al., 2015, although the majority of this work has focused on experiences related to specific identities (e.g., Flanders, 2015), or differences and similarities among different identities within the 2SLGBTQIA+ community (Fish et al., 2021; Meyer, 2003; Salerno et al., 2020). To our knowledge, only one study to date (Wardecker et al., 2022) has compared sexual minority to heterosexual adults on the experience of daily stressors and provided initial evidence that sexual minority adults may experience greater exposure to daily stressors than heterosexual adults. However, the methods used in this study, along with most work examining group differences in stressor exposure, used analytic methods that did not account for right-censored data and therefore makes the findings difficult to interpret. A thorough understanding of differences within and between various groupings of CH and SGM identities will move the field forward in understanding the risks and impacts of experiencing daily hassles that may have downstream impacts on mental health and well-being.

Given the growing literature identifying disparities between CH and SGM individuals in the experience of stress, including pandemic-related exacerbations (e.g., Hoyt et al., 2021), we were surprised that we did not find any differences between groups on the hazards of reporting stressors in daily life. There are several possible reasons for these null results. First, these null results could represent a true null effect or a Type II error, and it is impossible to determine which given the limitations of null hypothesis significance testing (Cohen, 1994). We also used a different analytic approach than the one study to date that showed differences between sexual minority and heterosexual adults on the frequency of stressor exposure (Wardecker et al., 2022). It is possible that accounting for right-censored data analytically could result in different patterns of significance than statistical approaches that compare mean frequencies between groups. Our sample was also different in terms of participant age, grouping of sexual and gender identity, period of data collection, geopolitical location, and pandemic context (discussed in more detail below), which all could contribute to the differences between our findings and the significant differences between groups observed by Wardecker and colleagues (2022). Regardless, it is crucial to account for right censoring when examining the risk of stressor exposure between groups so that accurate interpretations regarding whether and when SGM individuals may be at greater stressor-related risks than CH individuals can be made to appropriately target and scale intervention efforts. Inappropriate inferences can take needed and valuable time, care, and effort away from providing support to communities that need it. We are absolutely not suggesting that SGM individuals are not disadvantaged in any way but rather calling for greater methodological rigor in the study of daily events and experiences.

We also did not observe differences between SGM and CH individuals on stressor reactivity. Right censoring is not a concern in the analytic approach used to test this research question because the event data are the predictor, not the outcome variable. Thus, we cautiously suggest that our null findings, in contrast to Wardecker and colleague’s (2022) findings that sexual minority adults experienced greater stressor reactivity in daily life compared to heterosexual adults, could be related to some key differences between our samples. First, the two samples differ in the decades they were collected. Our data were collected between 2021 and 2022 from a university sample in Western Canada, whereas the data collection for their national US sample spanned from 1995 to 2014 (prior to the legalization of same-sex marriage in the US). Our data were also collected during the COVID-19 pandemic, whereas their sample was not, and we believe this context likely contributes in many ways to differences in the forms and possibly also the frequencies that stressors may present in daily life. We also had a different grouping of minority identity: The sample used by Wardecker and colleagues (2022) did not have information on gender beyond the male/female binary, whereas we did, so our categories of sexual/gender minority identity are different. There was also one key difference in the types of daily stressors that were measured: The measure of daily stressors used in the sample used by Wardecker and colleagues (2022) included an item regarding discrimination, whereas ours did not. Given that discrimination may be one key domain that differentiates the stressors experienced by SGM individuals, it is possible that we would have seen differences in stressor reactivity if we had measured discrimination as a type of daily stressor. In our results, SGM individuals reported greater negative mood in daily life than CH individuals and it is possible that stressors we did not measure, such as discrimination, contributed to this.

There are several other possible explanations for the null results regarding differences between SGM and CH groups related to our study design. It is possible that SGM individuals may have been exposed to higher levels of stressors than CH individuals pre-pandemic, and that CH individuals’ exposure to stressors may have increased to the level of SGM individuals during the pandemic. It was not possible for us to test this directly given that we did not have data prior to the onset of the pandemic. Another possibility is that the visibility of SGM participants’ identities could have more of an impact on stressor exposure and/or reactivity than their SGM identities alone. For example, an individual with an SGM identity who presents in a heteronormative way may be perceived by others as a member of the CH majority, and possibly be exposed to fewer stressors as a result. We did not measure SGM individuals’ outness or visibility in this study, and thus this is an important aspect of SGM individuals’ lived experiences to examine in future work. In addition, it is possible that for some types of stressors such as interpersonal conflicts, SGM individuals may be more likely to avoid conflicts in relationships than CH individuals given that SGM individuals may perceive their relationships to be more fragile than others given their minority identities. Taken together, our field has a long way to go in exploring the factors related to whether, when, and why SGM individuals may be at greater risk for exposure to some types of stressors and not others.

Although this was not a primary focus of our preregistered data analysis plan, we found that the hazards of some types of daily stressors showed much more variation between individuals than did others, and that some types of stressors had lower survival functions than others. There was no significant variation between individuals in the “any stressor” category and work/school related stressors. The survival functions for both categories (see Figure 3) show very low survival rates by the end of the sampling period. Specifically, for work/school stressors, on the first day of the observation period, just over 25% of the sample were “survivors” (had not experienced this type of stressor on the first day they reported), whereas by the end of the observation period, there were no survivors (i.e., everyone had experienced this type of stressor at least once). Similarly, for the “any stressor” event category, on the first day of the observation period, just under 25% of the sample were “survivors” (had not experienced this type of stressor on the first day they reported), whereas by the end of the observation period, there were no survivors (i.e., everyone had experienced this type of stressor at least once). These types of stressors may be so common in the student population that there is no variation in event hazard to explain. In contrast, we observed significant variation between individuals in the hazards of arguments, conflicts, disagreements; and family or home stressors, but the hazards of these stressors were not related to any predictors included in our models. There may be other individual differences beyond the demographics we examined that may explain who may be more at risk of these types of events than others, and more research is needed to understand what explains individual differences in the hazards of these specific types of daily stressors.

Although the strengths of this study include the sample size, intensive longitudinal assessment, and preregistered analyses, there are several limitations that should be kept in mind in interpreting these data. First, we did not collect information on where participants were located during the period of data collection for privacy reasons, and thus we do not know if participants were experiencing different types of provincial or federally regulated lockdown restrictions in each cohort if they were outside of British Columbia. Thus, the Cohort variable must be taken only as a proxy for pandemic context in terms of the shared experiences of university-level policies regarding online- or in-person classes, masking requirements, etc.

Second, the number of gender minority individuals (self-identified as a category outside of man or woman) in this study was low (n = 14 for the full sample), as was the number of individuals in each sexual minority identity category (see Table 1). These low numbers of separate identity categories prevented us from being able to examine unique experiences associated with each identity. Future research could over-sample specific sexual and gender minority identities to answer this specific question. In addition, our measure of gender identity included response options (among others) for “man”, “woman”, “trans man”, and “trans woman”, which may have been perceived by some participants as suggesting some forms of male-ness and female-ness are privileged over others. Inclusive questions related to gender and other identities are essential to preventing disenchantment or alienation participants might experience during their participation.

Third, our sample was homogeneous with respect to ethnicity (primarily White), student status (primarily domestic students), and gender (primarily individuals who identified as women). These demographics reflect what is typical for undergraduate student samples through Psychology departmental participant databases, but our study conclusions should be limited to individuals from this specific population as we cannot assume the results may generalize more broadly without further investigation.

Finally, as with any study examining the timing of events, it is possible that informative censoring was present in our data. Informative censoring refers to when there are censored cases or missing data on a time-to-event variable (e.g., daily conflict) that is related to some third, unknown variable (Allison, 1984). For example, it is possible that some participants may have chosen not to respond to the survey on a day when a high level of conflict was experienced. It is not possible to test informative censoring and care needs to be taken in the study design to prevent it (Allison, 2010). In this study, the low time required and broad response window of the daily surveys was likely to play a role in preventing informative censoring in the data, but as with any study of time-to-event data, it is important to factor the possibility of informative censoring into the interpretation of the study results.

Stressors in daily life are frequent, common, and important for long-term health and well-being. They may also be related to the mental and physical health disparities observed between SGM and CH individuals (Hoyt et al., 2021; Wardecker et al., 2022). More work is needed to examine whether, when, and why individuals with different identities may be at risk for encountering specific types of stressors in daily life using methods such as MSA that accurately account for right censored observations. Doing so will help us better understand how to tailor intervention efforts to reduce stressor exposure to specific populations who may be at particular risk.

The authors declare no conflicts of interest.

Contributed to conception and design: JPL

Contributed to acquisition of data: JPL, GK, SM

Contributed to analysis and interpretation of data: JPL, GK

Drafted and/or revised the article: JPL, GK, SM

Approved the submitted version for publication: JPL, GK, SM

This work was supported by funding awarded to Jessica P. Lougheed from the Aspire/Start Up Funding from the UBCO Irving K. Barber Faculty of Arts and Social Sciences (F20-03321).

We thank Alyssa Truong, Summer Fulford, Keaghan Forster, and Sierra Adamow-Boudreau for their assistance with data collection.

The pre-registration and all associated data analysis scripts and files are available on the OSF site. The pre-registration is available here: https://doi.org/10.17605/OSF.IO/P3FGD. All data analysis scripts and files are available here: https://osf.io/wvf2q/?view_only=efa6​d4a​999​684​0e9​b33​389​cc0​2f16​686