Introduction
Adverse life events (e.g. bankruptcy) are known to impact mental health negatively. It therefore may follow that accumulating adverse life events have cumulatively greater negative impacts on mental health. This is seen empirically with studies demonstrating a relationship between more adverse life events in childhood (Hughes et al., Reference Hughes, Bellis, Hardcastle, Sethi, Butchart, Mikton, Jones and Dunne2017) or throughout life (Richardson, Carr, Netuveli, & Sacker, Reference Richardson, Carr, Netuveli and Sacker2023) and worse mental health (e.g. higher risk of future psychiatric disorder). Further, and self-evidently, certain types of adverse events can impact mental health more than others (Kettlewell et al., Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020). There is less consistency in behavioural science: repeated exposure to adverse events in experimental settings may result in a sensitised (increased) response, or habituation (decreased) response (Blumstein, Reference Blumstein2016; Grissom & Bhatnagar, Reference Grissom and Bhatnagar2009; Rankin et al., Reference Rankin, Abrams, Barry, Bhatnagar, Clayton, Colombo, Coppola, Geyer, Glanzman, Marsland, McSweeney, Wilson, Wu and Thompson2009). This experimental disparity may manifest in different real-world impacts: Is the relationship between the number of adverse life events and mental health linear (each successive event has a similar effect on mental health)? Is it ‘sensitising’ (each successive event has a larger impact on mental health) or ‘desensitising’ (each successive event has a smaller impact on mental health)?
Theoretical models propose competing hypotheses. For example, the ‘stress sensitisation model’ (Stroud, Reference Stroud, Harkness and Hayden2020, Munroe & Harkness, Reference Monroe and Harkness2005) posits that the threshold of stress required to negatively impact mental health lowers with repeated exposure. Alternatively, repeated stress exposure may build tolerance (actively via new skills/strategies or passively via experience) and raise the threshold required for an event to affect mental health, e.g. ‘steeling hypothesis’ (Klopack, Wickrama, & Simons, Reference Klopack, Wickrama and Simons2022), ‘stress inoculation hypothesis’ (Lyons, Parker, Katz, & Schatzberg, Reference Lyons, Parker, Katz and Schatzberg2009), and ‘post-traumatic growth’ (Jayawickreme et al., Reference Jayawickreme, Infurna, Alajak, Blackie, Chopik, Chung, Dorfman, Fleeson, Forgeard, Frazier, Furr, Grossmann, Heller, Laceulle, Lucas, Luhmann, Luong, Meijer, McLean and Zonneveld2021). Such models imply that greater experience with adverse life events may reduce the negative mental health impact of future adverse life events. Seery (Reference Seery2011) also suggests ‘U-shaped’ relationships between lifetime exposure to adversity and mental health may exist, where people with some lifetime adversity may experience better mental health and wellbeing, and are less affected by later adverse events compared to people with high lifetime adversity or none.
Beyond early life events, there is limited empirical literature describing how repeated adverse adult life events affect mental health compared to a prior event. Kok et al. (Reference Kok, Twisk, Blom, Beekman and Huisman2021) used the Longitudinal Aging Study Amsterdam (adults 55–84 years in 1992/93), investigating whether there was a ‘sensitising’ or ‘steeling’ effect of cumulative negative life events on the mental health impact of proximate events over 15 years. Kok et al. (Reference Kok, Twisk, Blom, Beekman and Huisman2021) concluded that accumulation of negative life events reduced the impact of subsequent events on depressive symptoms, but ‘steeling’ was not completely protective – more events were associated with more depressive symptoms in older age. Seery, Holman, and Silver (Reference Seery, Holman and Silver2010) used an American online panel with four waves (2001–2004) to analyse the relationship between accumulation of 37 different adverse life events and mental health, finding that people with a history of some lifetime adversity reported better mental health outcomes than people with no or high history of adversity.
Furthermore, studies analysing cumulative adverse life events generally do not differentiate between event type, considering all events as equivalent exposures with equivalent impacts on mental health. This is unlikely to be true at face value. Kettlewell et al. (Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020) showed that life events differed in 1) severity of impact, 2) trajectory of recovery, and 3) effects on types of wellbeing. However, Kettlewell et al. (Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020) only studied the mental health impact of a single event (averaging effects of repeated life events of the same type).
Few studies have examined mental health impacts of specific repeated adverse life events compared to previous events. Both repeated climate change disasters (Li & Leppold, Reference Li and Leppold2025; Mitchell, Maheen, & Bowen, Reference Mitchell, Maheen and Bowen2024) and unemployment (Booker & Sacker, Reference Booker and Sacker2012; Gallo, Bradley, Teng, & Kasl, Reference Gallo, Bradley, Teng and Kasl2006; Watson & Osberg, Reference Watson and Osberg2017) may be sensitising. The comparison with Kok et al. (Reference Kok, Twisk, Blom, Beekman and Huisman2021) and Seery et al. (Reference Seery, Holman and Silver2010) suggests that the association between repeated negative life events and subsequent mental health may partly depend upon whether people experience a similar event repeatedly or several different events.
Using an adult population-based sample (Household, Income and Labour Dynamics in Australia (HILDA)), we aimed to evaluate the marginal mental health impact of seven common adverse life events, examining whether repeated exposure to the same event type was associated with sensitisation or habituation.
Methods
Data
This study used the HILDA Survey, a nationally representative longitudinal panel study of Australian households since 2001, starting with 13,969 persons in 7,682 households. This survey collects data (face-to-face interviews and self-completion questionnaires) annually from all people over 15 years in each household, covering social, health, and economic aspects of life (among others). Our study used Waves 1–23 (2001–2023).
Life events
In HILDA, life events have been recorded since Wave 2 (2002). This restricted our event sample for analysis from 2002 to 2023. Respondents report whether any of 22 major life events have occurred in the last 12 months, 13 of which are negatively valenced. Recording of life events is completed after the SF-36 questionnaire (from which we derived mental health scores), minimising the potential effect of the event (recall bias) on self-evaluation of mental health. Following Kettlewell et al. (Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020) we selected seven common adverse life events with the strongest negative effects on affective wellbeing (adjusted for age, education, co-occurring events and Socio-Economic Indexes for Areas). Similar to Kettlewell et al. (Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020) we excluded home destroyed by natural disaster, relative died, jailed, and a relative jailed, as these events were experienced by a relatively small number of individuals or were not measured for the entire observation period. Additionally, we chose to exclude widowed (death of spouse/child) and ‘Friend died’, as their conditional total impact on wellbeing was not significantly different from zero (Kettlewell et al., Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020).
Selected adverse life events for our analysis included: Fired or made redundant by an employer; Serious injury/illness to a close family member; Victim of physical violence; Victim of a property crime (e.g. theft); Separated/Divorced from a spouse/long term partner; Serious personal injury/illness (e.g. disability); Major worsening in financial situation (e.g. bankruptcy). Due to the known differences in severity of impact between event types (Kettlewell et al., Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020), each exposure was categorised by the cumulative count of reported events of that type. Categories were restricted to the first, second, and three or more events (four or more events of any one type were rare, Supplementary Figure S1).
Mental health
Mental health was measured using the Mental Health Inventory-5 (MHI-5), a validated five-item sub scale derived from the SF-36 measuring symptoms of depression and anxiety (Cuijpers et al., Reference Cuijpers, Smits, Donker, ten Have and de Graaf2009). Scores were rescaled from 0 to 100. Higher scores indicate better mental health. MHI-5 data from 2001 to 2023 were used.
Analysis
The analytic sample included respondents if adults (≥18 years) and had at least one adverse life event during the observation period (2002–2023). Events were included if MHI-5 scores were available the year before and after. We included non-event years with MHI-5 observations to build each participant’s baseline trajectory in mental health into the model.
Descriptive analysis compared baseline demographic characteristics (in the year of the first adverse event) between people with different numbers of total events to identify population heterogeneity.
We examined within-person change in mental health associated with cumulative adverse life events of the same type following a similar approach to (Ervin, Taouk, Hewitt, & King, Reference Ervin, Taouk, Hewitt and King2023). We modelled adverse event types separately, using a linear mixed-effects model with MHI-5 as the outcome and random individual intercepts. Each model included fixed effects for time, adverse event in the current year, cumulative event exposure (zero, one, two, three, or more) and an interaction between current year event occurrence and cumulative exposure. The interaction term tested whether the mental health impact of repeated adverse events (two, three, or more) was significantly different to the first event. Post hoc pairwise comparisons of estimated marginal means tested whether the second and third or more events had different mental health impacts.
We evaluated model construction using likelihood ratios (Supplementary Table S1). While including current adverse events consistently improved model fit across all exposures, improvements associated with cumulative exposure and interaction terms varied by event type. The final model specification was retained for consistency across event types. Further details on model definition are in the Supplementary Material.
We defined an adverse life event as ‘desensitising’ if the marginal effect of the second and third or more repeated events were smaller (less negative) than the first event; ‘unclear’ if the marginal effects of second and third or more event differed with respect to the first event (e.g. second event marginal effect less negative, but third or more event more negative); and ‘sensitising’ if both the second and third or more events had greater negative marginal effects than the first event.
Effect sizes were estimated using Cohen’s d. Simulation-based post hoc power analyses estimated power for detecting the effects of repeated event exposure on mental health (Green & Macleod, Reference Green and Macleod2016). Analysis was completed in R (version 4.3.1), using the lme4 package (version 1.1.34) for random effects modelling.
Sensitivity analysis
Our sensitivity analysis considered potential confounders. First, we added 1-year lagged MHI-5 to our base model to adjust for pre-event mental health. Second, we included a Mundlak correction (person-level mean adverse life events of that type over the observed period) in our base model. This adjusts for time-invariant unobserved heterogeneity (between-person differences in likelihood to experience adverse events) that may confound within-person (fixed) effects of repeated events on mental health (Mundlak, Reference Mundlak1978). Finally, we considered that consecutive events may be the same event over 2 years; meaning that a repeated event’s impact may be artificially lower if the starting mental health after the ‘first’ reported event is lower. To address this we ran the base model on our sample excluding consecutive events.
Results
Adverse life events
Our event sample (2002–2023) had 35,594 participants ≥18 years, with 23,109 adults experiencing at least one of the seven adverse life events in our study. Table 1 presents the number of each event type, and numbers and proportions of participants reporting one, two, or three or more events. The most common events were serious injuries/illnesses to family or personally (45% and 33% respectively); the least common was physical violence (8%). All other events were reported by at least 14% of respondents.
Adverse life events in HILDA (2002–2023), for an adult population (≥18 years of age)
Table 1. Long description
The table contains six columns: Event type, Adults with events N (%), Total events N, Adults with only one event N (%), Adults with total of two events N (%), and Adults with total of 3 plus events N (%).
* Major worsening in financial situation: 4,964 (14%) adults; 8,501 total events; 3,272 (66%) had one event; 901 (18%) had two; 791 (16%) had 3 plus.
* Separated or divorced from spouse or long-term partner: 6,508 (18%) adults; 10,804 total events; 4,056 (62%) had one event; 1,460 (22%) had two; 992 (15%) had 3 plus.
* Serious personal injury or illness: 11,706 (33%) adults; 26,247 total events; 5,832 (50%) had one event; 2,601 (22%) had two; 3,273 (28%) had 3 plus.
* Serious injury or illness to a close family member: 16,003 (45%) adults; 43,368 total events; 6,456 (40%) had one event; 3,468 (22%) had two; 6,079 (38%) had 3 plus.
* Victim of physical violence: 2,673 (8%) adults; 4,132 total events; 1,901 (71%) had one event; 441 (16%) had two; 331 (12%) had 3 plus.
* Victim of property crime: 6,972 (20%) adults; 10,867 total events; 4,632 (66%) had one event; 1,464 (21%) had two; 876 (13%) had 3 plus.
* Fired or made redundant by an employer: 5,818 (16%) adults; 9,014 total events; 3,889 (67%) had one event; 1,223 (21%) had two; 706 (12%) had 3 plus.
a HILDA has a total of 35,594 unique adult participants in the dataset (2002–2023).
Table 2 presents demographic characteristics of the analytic sample at the first event, stratified by total number of repeated events. People who experienced more repeated events tended to be older at the first event for major worsening in finances, serious personal injury/illness, and injury/illness to family; and younger for victims of physical violence/property crime. There was no clear difference between people with different total events in the age of first event for separation/divorce or firing/redundancy. People who experienced more repeated events tended to be female at the first event for separation/divorce, serious injury/illness to family, and victim of physical violence; and male for firing/redundancy. Socioeconomic disadvantage (unemployment, lower income, lower education level, and CALD background) was associated with a greater number of repeated adverse life events in 17 of the 35 associations evaluated. Trends for baseline unemployment were largely statistically insignificant, but there were statistically significant differences at the first event between people with different total events for serious personal injury/illness (no clear trend) and serious injury/illness to family (people with more events tended to have higher rates of unemployment at the first event). Annual household income tended to be lower at baseline for people who would experience more repeated adverse events, and this was a statistically significant difference for all seven event types, including major worsening in finances. A CALD background was associated with lower rates of repeated serious injury/illness to a close family member and victim of physical violence but higher rates of repeated separation/divorce.
Demographic characteristics of participants (≥18 years of age) who experienced any adverse life events, at the time of the first event (2002–2023)
Table 2. Long description
The table is divided into seven sections based on the type of adverse life event. Each section lists characteristics: Age, Male percentage, Unemployed percentage, Household income per $10,000, Tertiary education percentage, Did not finish high school percentage, and C A L D percentage.
1. Major worsening in finances: N equals 3,272 for 1 event, 901 for 2 events, and 791 for 3 plus events. Household income significantly decreases as events increase (7.9 to 6.0, p-value less than 0.001).
2. Separation or divorce: N equals 4,056 for 1 event, 1,460 for 2 events, and 992 for 3 plus events. Significant trends include a decrease in male percentage (47 to 42 percent, p-value 0.024) and household income (7.8 to 6.0, p-value less than 0.001) as events increase.
3. Serious personal injury or illness: N equals 5,832 for 1 event, 2,601 for 2 events, and 3,273 for 3 plus events. Age significantly increases with event frequency (46 to 51 years, p-value less than 0.001).
4. Serious injury or illness to a close family member: N equals 6,456 for 1 event, 3,468 for 2 events, and 6,079 for 3 plus events. Male percentage significantly drops from 48 to 39 percent (p-value less than 0.001) as events increase.
5. Victim of physical violence: N equals 1,901 for 1 event, 441 for 2 events, and 331 for 3 plus events. Male percentage (49 to 35 percent) and C A L D status (8.6 to 3.3 percent) both show significant decreases (p-value less than 0.001).
6. Victim of property crime: N equals 4,632 for 1 event, 1,464 for 2 events, and 876 for 3 plus events. Household income significantly decreases from 8.5 to 6.9 (p-value less than 0.001).
7. Fired or made redundant: N equals 3,889 for 1 event, 1,223 for 2 events, and 706 for 3 plus events. Male percentage significantly increases from 54 to 67 percent (p-value less than 0.001) as event frequency rises.
CALD = Culturally and Linguistically Diverse, represents people born in a non-English-speaking country (United Kingdom, New Zealand, Canada, USA, Ireland, and South Africa) OR did not grow up speaking English as their first language.
a Mean reported for continuous variables; Percentages for categorical variables.
b CI = confidence interval.
c P-values derived from Kruskal-Wallis rank sum test (continuous variables); Pearson’s Chi-squared test (categorical variables).
Table 3 reports linear mixed effects regressions run for each event type, examining the relationship between repeated adverse events and mental health. This includes Model 1 (base model), Model 2 (adjusted for pre-event mental health [1-year lagged MHI-5]) and Model 3 (adjusted for person-level mean adverse events of that type over the observation period [Mundlak correction]).
Linear mixed-effects modelling examining the association between cumulative adverse life events and MHI-5 in adults (>18 years) in HILDA (2002–2023) for different event types
Table 3. Long description
The table presents regression results on current-year M H I 5 scores for seven categories of adverse life events. Each category includes data for 1st event, 2nd event total effect, Delta versus 1st event, 3 plus event total effect, Delta versus 1st event, and Delta versus 2nd event.
* Financial worsening: 1st event coefficients range from minus 5.97 to minus 6.95. 2nd event total effects range from minus 6.85 to minus 8.06.
* Victim of property crime: 1st event coefficients range from minus 0.68 to minus 1.09. 3 plus event total effects range from minus 1.73 to minus 1.92.
* Victim of physical violence: 1st event coefficients range from minus 4.65 to minus 5.33. 2nd event total effects range from minus 5.77 to minus 6.07.
* Separation or divorce: 1st event coefficients range from minus 4.56 to minus 5.11. 3 plus event total effects range from minus 2.71 to minus 3.59.
* Serious personal injury or illness: 1st event coefficients range from minus 4.33 to minus 4.78.
* Serious family injury or illness: 1st event coefficients range from minus 1.57 to minus 1.87.
* Fired or redundancy: 1st event coefficients range from minus 1.62 to minus 2.02.
Model 1 is the base model. Model 2 adjusts for baseline mental health. Model 3 includes the Mundlak correction. Significance levels are indicated by asterisks where 3 asterisks mean p is less than 0.001.
Model 1: Base model adjusted for year, current adverse event, cumulative adverse events and interaction between current and cumulative adverse events.
Model 2: Model 1 adjusted for baseline mental health (1-year lagged MHI-5).
Model 3: Model 1 adjusted for Mundlak correction (person-level mean adverse events of that type in the observed period).
*p < 0.05; **p < 0.01; ***p < 0.001.
a Total effects at 2nd and 3+ event derived from post hoc estimated marginal means (tests whether total effect is different to zero). Standard errors computed via delta method from model variance–covariance matrix.
b Δ vs 1st event refers to the Model interaction term. (additional effect relative to 1st event).
c Δ vs 2nd event refers to post hoc pairwise contrasts to test whether marginal effects (2nd vs 3+ event) were significantly different. Standard errors computed via delta method from model variance–covariance matrix.
At the first occurrence, all adverse events were associated with a decrease in MHI-5, with financial worsening having the largest effect and being a victim of property crime having the smallest effect. Adjusting for pre-event mental health (Model 2) and a Mundlak correction (Model 3) reduced the negative effect of a first adverse event on MHI-5. For two and three or more repeated adverse events, marginal effects analysis showed that event occurrence (regardless of cumulative exposure) generally had a statistically significant negative impact on mental health. However, the mental health impact of repeated adverse events (relative to a previous event) differed by event type.
Results for two adverse event types were suggestive of sensitisation. For financial worsening, there was a statistically significant greater mental health impact from the second and third or more event compared to the first. For property crime, only the third or more occurrence showed a statistically significantly greater impact than the first, while the second did not.
Converselyesults for two other adverse event types were suggestive of habituation. Repeated separation/divorce and personal injury/illness resulted in smaller negative impacts on mental health compared to the first occurrence; however, this was only statistically significant for personal injury/illness in Model 2.
Repeated physical violence showed a potential ‘U-shaped’ relationship on average (second event worse than first, third or more event less worse than the second and similar to first); however, this was not statistically significant. Repeated serious family injury/illness and firing/redundancy showed no statistically significant changes in mental health compared to the first event.
Post hoc pairwise contrasts comparing the mental health impact of the second vs third or more adverse event did not show statistically significant differences for any event type.
Adjusting for baseline mental health (Model 2) made interaction coefficients more positive, indicating that the mental health impact of repeated events may be attenuated by pre-event mental health. A Mundlak correction (Model 3) made interaction coefficients for repeated events more negative, suggesting that between-person effects (individuals who experience repeated adverse events tend to have poorer mental health) may suppress the within-person effect of repeated events in the base model (Model 1). Direction of coefficients was largely consistent across Models 1–3. Total effects of cumulative events predicting one-year MHI-5 change are shown in Figure 1 for Models 1–3.
Marginal effects of repeated adverse events on MHI-5 score, for different event types and model specification.
Figure 1. Long description
The figure consists of seven panels arranged in a three-by-three grid with the last two slots empty. The y-axis for all panels represents the 1 Year Change in M H I-5 Score, with values generally ranging from 0 down to -8. The x-axis represents the number of Repeated Events categorized as 1, 2, or 3 plus.
Three model specifications are plotted for each event count:
1. Model 1 (Base) shown as a black triangle.
2. Model 2 (Base plus Baseline Mental Health) shown as a black square.
3. Model 3 (Base plus Mundlak Correction) shown as a black circle.
Panel 1: Major worsening in finances. Shows the largest negative impact, with scores between -6.0 and -8.1. The negative effect increases slightly from the first to the second event.
Panel 2: Separation or divorce from a spouse or long-term partner. Scores range from -2.7 to -5.1. The negative impact appears to lessen slightly as events repeat.
Panel 3: Serious personal injury or illness. Scores range from -3.5 to -4.8, showing a relatively stable negative effect across repetitions.
Panel 4: Serious injury or illness to a close family member. Scores range from -1.5 to -1.9, showing the smallest negative impact among the first four panels.
Panel 5: Victim of physical violence. Scores range from -4.5 to -6.5. The second event shows a deeper negative peak compared to the first and third plus events.
Panel 6: Victim of property crime. Scores range from -0.7 to -2.1, showing high variability with large 95% confidence interval error bars.
Panel 7: Fired or made redundant. Scores range from -1.3 to -2.3, with the negative impact increasing at the 3 plus event mark.
In most panels, Model 2 (square) tends to show a slightly less negative marginal effect than Model 1 (triangle) and Model 3 (circle). All data points include vertical error bars representing 95% confidence intervals.
Supplementary Material reports coefficients for other variables (Supplementary Table S2). Greater cumulative exposure to adverse events was generally associated with lower MHI-5, suggesting a lasting effect of prior adverse experiences on mental health beyond current event years and in non-event years. This was consistent across all event types except separation/divorce, where cumulative exposure showed small positive coefficients. For event types where cumulative exposure effects were consistently statistically significant in the base model (all except property crime and separation/divorce), coefficients attenuated towards zero after adjusting for prior mental health (Model 2) and a Mundlak correction (Model 3), suggesting effects are partly explained by prior mental health status and between-person effects.
The coefficient for 1-year lagged MHI-5 was positive for all events, indicating better pre-event mental health was associated with better current-year mental health. Finally, the Mundlak correction coefficient was negative for all events, indicating that people who experienced more adverse events were likely to have poorer mental health.
Supplementary Material includes further detail on model fit (Supplementary Tables S2 and S3). Compared to the base model (Model 1), there was a significant improvement in AIC, BIC, and log likelihood with baseline pre-event mental health (Model 2) and a smaller improvement from a Mundlak correction (Model 3). Low marginal R2 for Models 1 and 3 reflects that specific adverse events explain a small proportion of total variance in mental health. This improved significantly when adding pre-event mental health (Model 2). Higher conditional R2 in Models 1 and 3 reflects between-person variance captured by the random intercept, reduced in Model 2 when pre-event mental health is added. The intraclass correlation coefficient (ICC) of 0.54–0.59 in Models 1 and 3 indicates that roughly half of variance in MHI-5 is attributable to between-person differences. The attenuation of ICC in Model 2 (0.36–0.43) indicates that prior mental health accounts for some between-person differences.
Cohen’s d assessed effect sizes in Model 1 (Supplementary Table S5). Financial worsening had the largest Cohen’s d (~0.4 for first, second and third or more events), whereas experiencing property crime had the lowest (<0.11). Effect sizes were small (0.2–0.35) for experiencing property crime, serious personal injury/illness and separation/divorce. Effect sizes for serious family injury/illness and firing/redundancy were even smaller (0.1–0.2).
Simulation-based post hoc power analyses assessed Model 1’s ability to detect differences between second and third or more adverse life events compared to the first (Supplementary Table S6). Power varied significantly across event types and exposure contrasts. Adequate power was observed for some comparisons, most notably separation/divorce (three or more vs first), whereas power was moderate to low for other events, particularly higher-order exposure contrasts. This reflects variation in the number of individuals contributing to repeated exposure categories and heterogeneity in exposure trajectories across events. Hence, non-significant interaction effects in Table 3 should be interpreted as inconclusive rather than no effects.
Sensitivity analysis – excluding events reported in consecutive years
Running the base model (Model 1) on a sample excluding events in consecutive years changed the interpretation for several events. Compared to the full sample, this showed habituation to physical violence from the third or more event compared to the first event. Additionally, the third or more separation/divorce showed further desensitisation compared to the second event. In contrast, the third or more financial worsening and property crime no longer had statistically significant differences in their effect on MHI-5 compared to the first event. Other event types maintained the same patterns.
Discussion
Our work evaluates the marginal mental health impact of repeated exposure to the same type of adverse life event compared to a single event in a population-representative cohort. We found evidence of habituation for repeated separation/divorce from a long-term partner. Habituation was also observed on average for repeated serious personal injury/illness, but with less confidence. We found support for sensitisation in repeated financial worsening (two and three or more events) and property crime (three or more events). In our sensitivity analysis for consecutive events, we also found some evidence for habituation in physical violence (when experienced three or more times). There was no clear trend for being fired or made redundant, or serious injury/illness to family.
Interestingly, for those events which showed sensitisation and habituation, mental health responses tended to ‘plateau’ after the second exposure (that is second and third or more events were worse/better than the first, but not significantly different from each other). For some events, only the third or more event was worse/better than the first (and not the second), suggesting a potential ‘threshold’ effect’. Such concepts may be analogous to ‘stress-diathesis’ models (Monroe & Simons, Reference Monroe and Simons1991; Zubin & Spring, Reference Zubin and Spring1977) which propose that psychiatric disorder may arise once cumulative stress exceeds a person-specific ‘threshold’. Similarly, an individual’s response to adverse events may only change after a threshold (e.g. certain number of events) and stabilise afterwards.
More generally, our habituation findings may be consistent with Kok et al. (Reference Kok, Twisk, Blom, Beekman and Huisman2021) in an older adult Dutch cohort (55–84 years), which found that accumulation of previous negative life events reduced the impact of subsequent events on depressive symptoms. They examined the interaction of cumulative and new life events, finding lower depressive symptoms from new events (one, two and three or more); however, this was only statistically significant for people with two cumulative events. However, these results are not directly comparable. Their study was in an older cohort, with different event types, with their most common baseline events being bereavement (25%) and parental death (15%). Their most common proximate events were illness/death of a loved one (87%), with only 2% victims of crime and 0.1% divorce.
The literature has some theoretical explanations for observed desensitisation, including Dienstbier’s theory of psychophysiological toughness (Dienstbier, Reference Dienstbier1989), where episodes of manageable stress equip people to manage future stress more easily, reducing mental health impacts in those already ‘resilient’; the aforementioned ‘steeling hypothesis’ (Klopack et al., Reference Klopack, Wickrama and Simons2022), ‘stress inoculation hypothesis’ (Lyons et al., Reference Lyons, Parker, Katz and Schatzberg2009) and ‘post-traumatic growth’ (Jayawickreme et al., Reference Jayawickreme, Infurna, Alajak, Blackie, Chopik, Chung, Dorfman, Fleeson, Forgeard, Frazier, Furr, Grossmann, Heller, Laceulle, Lucas, Luhmann, Luong, Meijer, McLean and Zonneveld2021). Kok et al. (Reference Kok, Twisk, Blom, Beekman and Huisman2021) explored potential psychosocial mechanisms, finding that the desensitisation effect was smaller in people with a stronger internal locus of control and higher emotional stability.
There has been limited previous work analysing how the mental health impact of specific adverse life events varies with the number of repeated events, with no previous empirical studies to our knowledge examining repeated financial worsening, separation/divorce and personal injury/illness. Again, we found that the impact of adverse life events differs by type of event (Kettlewell et al., Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020), limiting inferences from empirical studies which combine events and challenging the generalisability of theoretical models.
For firing or redundancy, there is some analogous literature. Gallo et al. (Reference Gallo, Bradley, Teng and Kasl2006) analysed an American study of older workers (51–61 years old) with recurrent manufacturing job losses, finding a non-statistically significant increase in depressive symptoms associated with the second job loss compared to the first job loss. Additional involuntary unemployment (3rd and 4th job losses) was associated with a non-statistically significant decrease in depressive symptoms (Gallo et al., Reference Gallo, Bradley, Teng and Kasl2006). In contrast, we find repeated job losses had similar effects on mental health compared to the first job loss. However, our results were also not statistically significant. Another study of the British Household Panel Survey suggests that economically inactive people prior to unemployment report worse mental health with each unemployment episode, whereas people who are previously employed have stable mental health responses (Booker & Sacker, Reference Booker and Sacker2012). In combination with our results this suggests likely heterogeneity in the population of job losses mediating the impact on mental health.
Finally, property crime had the smallest effect size on mental health at the first event compared to the other events (Supplementary Table S5). This was a similar finding for the overall conditional effect of property crime on wellbeing in the Kettlewell et al. (Reference Kettlewell, Morris, Ho, Cobb-Clark, Cripps and Glozier2020) study. A previous HILDA study (Freeman & Smith, Reference Freeman and Smith2014) has shown that the negative impact of property crime on mental health (measured by SF-36 mental health scores) was small and not statistically significant, with violent crime having more significant negative effects (as we also found). The sensitisation effect on mental health for property crime was seen only after experiencing three or more events, which may suggest a threshold effect.
The ability to identify and understand if certain types of adverse life events are ‘sensitising’ or ‘habituating’ in their mental health impacts is of policy and clinical relevance, although care must be taken in making inferences. For instance, although the marginal impact of cumulative exposure to physical violence, the least common event, was not significantly different from the first, the effect size of repeated violence was one of the largest (Supplementary Table S4). Additionally in our sensitivity analysis excluding consecutive events, physical violence experienced three or more times may show habituation. Depending on worldview this might be explained by desensitisation or endogenous differences between people who experience violence. Notably, women were vastly overrepresented (64%) in those reporting three or more experiences of physical violence, indicating potential domestic violence. Habituation may reflect ‘emotional numbing’ seen in post-traumatic stress disorder (Duek, Seidemann, Pietrzak, & Harpaz-Rotem, Reference Duek, Seidemann, Pietrzak and Harpaz-Rotem2023), hardly a positive indicator. It may be overly simplistic to use this empirical data to support such hypotheses as ‘resilience’, ‘steeling’, or ‘post-traumatic growth’, as always being a ‘good thing’.
Limitations
We note several limitations. We are unable to measure or control for the ‘severity’ of the adverse event and have assumed each event is similarly ‘severe’. There is a subjective tautology as the definition of severity might include the psychological response we are using as the outcome. Additionally, we are unable to observe an individual’s response to the event, which may attenuate its mental health impact. For example, an individual may be more sensitive to repeated financial losses or redundancy due to prior experiences but has found additional employment before their next observation, mitigating its mental health impact.
People may also report events in different ways. For example, if an event occurs over 2 years, they may report the same event twice. Alternatively, repeated events in consecutive years (15–28% of events, depending on event type) may obfuscate the mental health impact of the consecutive event (potentially starting from a lower baseline mental health). We have attempted to address this by running the base model on a sample excluding consecutive events. We note that the study was underpowered for certain events (e.g. firing/redundancy, physical violence, and experiencing property crime twice) which may have resulted in non-significant findings (Supplementary Table S6).
Multiple events of the same type may occur in a single year, less likely for events such as divorce, but possible for others e.g. personal injury/illness; however, we are unable to control for this. We have not controlled for co-occurring events due to complexity in interpretation. We also do not know what is the true ‘first event’, only the ‘first event’ in HILDA. Similarly, missing life events may exist (survey non-completion or loss to follow up). However, bias should be limited given good follow up (78–84%) for people who experience an adverse life event (Supplementary Table S11), survey completion was generally high (>89% for events [Supplementary Table S12] and MHI-5 scores [Supplementary Table S13]), and low event frequency. The study methods and adult sample also mean we cannot address the interaction of life events with childhood adversity (Stroud Reference Stroud, Harkness and Hayden2020).
Finally, we acknowledge the likely role of population endogeneity influencing the relationship between repeated adverse life events and mental health (evident in demographic differences in Table 2). Additionally, people with more total events generally had lower baseline mental health at the first event (Supplementary Table S7). Nonetheless, the focus of our study was identifying within-person effects, and our sensitivity analysis with a Mundlak correction shows robustness when adjusting for time-invariant between-person confounding. Investigating between-person endogeneity is an avenue for future study to investigate how personal characteristics attenuate the impact of repeated adverse life events on mental health.
Conclusion
We provide a framework for modelling the marginal effects of repeated negatively valenced stimuli on mental health. We present empirical evidence of habituation for repeated separation/divorce, and sensitisation for repeated financial worsening and property crime. Future research should investigate heterogeneity and potential endogeneity associated with the effects of repeated events on mental health, and apply empirical findings to theories of psychological adaptation.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726105479.
Data availability statement
The data that support the findings of this study are available from Household, Income and Labour Dynamics in Australia Survey (HILDA) conducted by the Australian Government Department of Social Services (DSS), but restrictions apply to the public availability of this data, which was used under licence for the current study and so is not publicly available.
Acknowledgements
This paper uses unit record data from Household, Income and Labour Dynamics in Australia Survey [HILDA] conducted by the Australian Government Department of Social Services (DSS). The findings and views reported in this paper, however, are those of the authors and should not be attributed to the Australian Government, DSS, or any of DSS’ contractors or partners. DOI: 10.26193/NBTNMV
Author contributions
All authors contributed to study conception and design. Data analysis was performed by A.C. and R.M., with significant input, interpretation and supervision from N.G. The manuscript was drafted by A.C., and all authors edited previous versions of the manuscript. All authors read and approved the final manuscript.
Funding statement
This research was supported (partially or fully) by the Australian Government through the Australian Research Council’s Centre of Excellence for Children and Families over the Life Course (Project ID CE200100025).
Competing interests
The authors declare no competing or conflict of interests.
Ethical standards
This research was deemed negligible risk by the University of Sydney Human Research Ethics Committee and granted an ethics exemption on October 26th, 2023.
Consent for publication
This research uses unit record data from the Household, Income and Labour Dynamics in Australia Survey (HILDA) conducted and made available by the Australian Government Department of Social Services (DSS).
Open access
