Rural and Agricultural Natural Disaster Stress and Recovery: A Comparison

ABSTRACT This study used a novel survey instrument to evaluate the hypothesis that U.S. agricultural producers have significantly different stress and recovery experiences following acute-onset natural disaster compared to their non-agricultural counterparts. Participants were recruited through local organizations and targeted email and social media in communities in Arkansas and Nebraska that had experienced violent tornadoes in 2014 and/or severe flooding in 2019. The survey instrument incorporated the Brief Resilience Scale, the Revised Impact of Event Scale referencing two time points, the Posttraumatic Growth Inventory-Short Form, and original questions. Demographic, exposure, stress, and recovery measures were analyzed in SAS with Chi-square tests, t-tests, Wilcoxon tests, and multiple linear regression modeling to test for differences between agricultural and non-agricultural groups in resilience, event exposure, stress symptoms in the week after the event, stress symptoms in the month before the survey, a calculated recovery ratio, and posttraumatic growth. Analysis sample (N = 159) contained 20.8% agricultural occupation, 71.1% female, and 49.1% over age 55. No significant differences were found between agricultural and non-agricultural participants when comparing resilience, stress, or recovery ratio measures. Unadjusted posttraumatic growth score was significantly lower in the agriculture group (P = .02), and an occupation group by sex interaction was significantly associated with posttraumatic growth score (P = .02) when controlled for number of initial posttraumatic stress symptoms in the adjusted model, with agricultural women showing lower growth. Overall, there was no evidence of significant difference in disaster stress and recovery between agricultural and rural, non-agricultural groups in this study. There was some evidence that women in agriculture may have lower levels of recovery. Data indicated that rural residents continue to experience posttraumatic-type symptoms up to 8 years beyond the acute-onset natural disaster events. Communities should include strategies to support mental and emotional health in their preparedness, response, and recovery plans with intentional inclusion of agricultural populations.


Introduction
Living through a natural disaster can range from inconvenient or disruptive to terrifying, traumatic, and life changing.Since 1980, 355 natural disasters individually exceeding a billion dollars in cost (2023 cost-adjusted) and together exceeding 2.5 trillion dollars, not to mention many less costly events, have occurred in the U.S., affecting all 50 states. 1Even more concerning is the increasing frequency of these devastating events.The average number of billion-dollar weather disasters per decade in the U.S. has steadily risen from 3.3 per year in the 1980s (average annual total $21.0 billion) to 13.1 in the 2010s ($95.0 billion).The last 3 years (2020-2022) averaged 20 events per year ($149.2 billion annually). 1Extreme weather impacts property, infrastructure, and health with individual effects such as posttraumatic stress, depression, and substance use. 2 Because of the connection between agriculture and weather, it is important to understand whether agricultural populations have unique risk or resilience affecting mental and emotional health status when faced with natural disaster.
In 2019, 3.6 million people in the U.S., or 1.8% of the workforce, were directly employed in farming, forestry, and fishing activities. 3Existing research highlights potential mental health effects for disaster survivors, 2 but little is known regarding experiences specific to U.S. agricultural producers faced with acute-onset natural disasters such as floods and tornadoes.
An extensive literature review provided limited information on natural disaster stress and recovery in U.S. agricultural populations.Greater depressive symptoms were reported in residents of small towns and rural communities than in larger cities or farm populations following 1993 Midwest floods. 4Fishing captains in the Northeast experienced increased psychological distress following widespread fishery failure, a slow-onset disaster. 5esearch has been conducted examining the mental health needs of veterinarians following natural disasters and the potential to employ this group as a mental health resource in disaster-affected agricultural settings. 6Increased occupational stress in agricultural producers has been associated with drought during the growing season. 7][10][11][12] Although some research on international agricultural populations has been published, primarily related to drought in Australia, conclusions may not be generalizable to U.S. populations due to differences in culture, social networks, or community infrastructure.In addition, analysis from slow-onset disasters like drought may not generalize to acute-onset events.
The Central States Center for Agricultural Safety and Health (CS-CASH) in Omaha, Nebraska, supported the Rural Natural Disaster Stress and Recovery (RNDSR) survey development and study.This research aimed to assess disaster mental health experiences related to acute-onset natural disasters in U.S. rural and agricultural populations to provide evidence for disaster preparedness, response, and recovery to support mental and emotional health.We hypothesized that agricultural producers have different disaster stress and recovery experiences compared to rural nonagricultural counterparts.Anecdotally, agricultural producers -farmers, ranchers, and fishers -have a reputation for strength and resilience, so we might expect them to experience less stress and greater recovery. 13,14However, they have a particular dependence on weather along with an elevated suicide rate, 15 so we also consider the possibility of greater disaster mental health risk.

Human subjects protection
The University of Nebraska Medical Center (UNMC) Office of Regulatory Affairs approved the Rural Disaster Stress and Recovery Study as exempt research under IRB #729-21-EX.While human subjects were involved in the survey study, no identifying protected health information was collected.

Study design
We collected and analyzed survey data from agricultural and non-agricultural populations in targeted disaster-affected rural communities in a cross-sectional observational study of a voluntary convenience sample.Only acuteonset natural disaster events such as flood, tornado, or fire were included.Slow-onset natural disasters such as drought, manmade disasters such as chemical accident or war, and disease outbreaks such as COVID-19 were excluded.
An a priori sample size estimate of 128 was calculated (G*Power 3.1.96)based on alpha = 0.05, power = 0.80, and a medium effect size d = 0.5 in a t-test for two independent samples with allocation ratio 1:1 between agricultural and nonagricultural subgroups, allowing for 20% incomplete surveys.

Study populations
Agricultural and non-agricultural populations affected by natural disasters in Arkansas and Nebraska were targeted for participation in this research, based on all authors' personal knowledge, relationships, and networking in impacted communities.On April 27, 2014, an EF-4 tornado (winds 166-200 mph) struck Mayflower (population 1,984) 16 and Vilonia (4,288) 16 in Faulkner County, Arkansas, with 16 fatalities, 17 and 400-500 homes were destroyed along a 41mile path. 18This county and others were also affected by destructive Arkansas River flooding in May-June 2019, part of a $3.3 billion event. 1 In Nebraska, the town of Pilger (240) 16 and the surrounding region were struck by two EF-4 tornadoes on June 16, 2014, with 20 injuries and 2 fatalities; 19 other tornadoes occurred nearby during the same outbreak.The town and vicinity of North Bend (1,279) 16 experienced significant damage to infrastructure, property, and agricultural operations from Platte River flooding in March 2019, a disaster that cost over $11 billion. 1 The RNDSR survey was delivered in English via targeted communities' local social media pages, email and postal mail to agricultural organization member lists, personal contact in local communication hubs, agricultural newsletters and social media, and local media reports.CS-CASH collaborated with community partners, including extension agents, farm association representatives, health department official, newspaper manager, and community center director.Study data were collected from December 2021 through February 2022 and managed using the REDCap electronic data capture tools hosted at the University of Nebraska Medical Center (UNMC).Service and support was provided by the Research Information Technology Office (RITO), which is funded by the UNMC Vice Chancellor for Research.

Survey components
Validated survey instruments measuring resilience, stress, and recovery that had been tested in diverse and disaster-affected populations were adapted for the RNDSR survey (Appendix A).The Brief Resilience Scale (BRS) was designed and validated to "assess resilience as bouncing back from stress." 20Respondents indicated their level of agreement with six statements about their typical responses to stressful events.The scale was scored by averaging values of 1-5 assigned to Likert-type responses.Three of the six items were reverse scored.
The Revised Impact of Event Scale (IES) 21 is a screening tool for posttraumatic stress disorder (PTSD) but was used here as a tool for counting commonly experienced intrusion (re-experiencing) and avoidance posttraumatic stress symptoms without accounting for frequency or intensity of those symptoms.Minor modifications to the survey instructions were made, but no changes were made to questions.This scale has been used and tested in diverse populations after stressful events, including post-disaster, and was designed for use at any length of time after a stressful event. 21ubjects completed the Revised IES based on their memory of presence or absence of 15 possible posttraumatic stress symptoms in the first 7 days following their primary disaster event. 22ubjects were also given the option to select Don't recall for each symptom.The Revised IES was scored here as number of symptoms reported.
Subjects then completed the Revised IES scale again regarding the same 15 possible event-related symptoms occurring in the 30 days prior to completing the survey.The time difference of 30 days for the present scale versus 7 days for the past allowed for the expected trend of reduced symptoms over time while capturing symptoms still experienced even if less frequent.
The Posttraumatic Growth Inventory -Short Form (PTGI-SF) 22 is a self-report measure of recovery as positive personal growth rather than reduced posttraumatic stress symptoms.Subjects chose the degree to which 10 specific positive changes occurred in their life due to the stressful event, from Not at all to Very great degree (six options scored 0-5).The PTGI-SF score was the sum of responses, ranging from 0 to 50.
Demographic data were collected.Age, sex, race or ethnicity, rural or urban residence, specific disaster event, agricultural occupation, and presence of dependents in the home were independent categorical variables.Rural was defined as residence in a location of population less than 10,000, following the U.S. Census Bureau's proposed definition based on the 2020 U.S. Census. 23Choices for primary occupation were Farm, Ranch, Fishery, and Not in Agriculture, and the first three were combined into a single Agriculture occupational group in data preparation due to sample size limitations.
No distinction was made between farmworker and farm owner/operator.
Exposure questions evaluated direct or indirect impact, property loss, displacement, financial hardship, injury to self or family member, and fear for life of self or family member.These were summed for a single exposure score where perceived direct impact was 2 points, indirect was 1, no impact was 0, and all other exposures were 1 point for presence or 0 for absence.This method combining perceptions and objective experiences was adapted from the literature. 9,11Qualitative open field responses and a novel Resource Use and Effect inventory to be described in a separate report were also collected.

Statistical analysis
Statistical analysis was performed in SAS Studio 3.8 (Enterprise Edition) (Cary, NC, USA).Descriptive statistics and graphs were produced to review population characteristics and scale score distributions.Chi-square tests for equal proportion and t-tests for independent means were conducted to identify significant differences between agricultural and non-agricultural occupational groups.Wilcoxon non-parametric tests were also used for confirmation where outcomes and residuals were not normally distributed.Linear and multiple regressions were used to test association of occupational group with outcome scores while controlling for covariates, including age group, sex, race or ethnicity, disaster type, exposure level, dependents in home and years since event.An additional outcome Recovery Ratio (RR), the proportional reduction in symptom count from time of event to present was calculated as IESpastÀ IESnow IESpast and tested with the same procedures.A significance level of α = 0.05 was used for all hypothesis testing.Across all comparisons, post hoc power was 0.94-0.98 for detecting a large effect, and 0.59-0.71for medium effect.

Sample size and population
The full data set included 216 records, which was reduced to 159 for analysis.Subjects were flagged for completeness of each scale, defined as at least 5/6 exposure questions completed, 5/6 BRS, 13/15 Revised IES, and 8/10 PTGI-SF.Only individuals with completed scales were included at each stage  of the analysis.Sample size completing the last scale was 126.See Figure 1 for sample size flow.See Table 1 for demographic and disaster event details.The proportion of male to female was significantly different in the agricultural group (17-16)  compared to the non-agricultural group (23-93) on chi-square test (P < .001);therefore, sex was controlled for in further occupational group analysis.

Outcomes
The agricultural and non-agricultural occupational groups were compared on each scale covering an aspect of a stress and recovery cycle (Figure 2), where each statistical null hypothesis was for no difference in mean or median score by occupational group.

Resilience
On analysis of covariance (ANCOVA), the difference in mean BRS by occupational group was not statistically significant when controlled for sex (P = .135).Age group, race or ethnicity, event state, dependents in home, and presence or absence of disaster event were not significant as covariates.

Event exposure
Exposure scores were calculated for 146 subjects; 122 (83.6%) reported, in their opinion, being directly affected by the disaster, and 24 (16.4%)indirectly.There were 82 (56.6%) who reported losing property, 75 (51.4%)were displaced from their home, 69 (47.9%) experienced financial hardship due to the event, 3 (2.1%) reported injury to self or family member, and 78 (53.4%) feared for their life or a family members.Overall distribution of exposure score was approximately normal (mean 3.94, CI 3.69-4.19), but the agricultural subgroup (N = 28) showed a more uniform distribution.Agriculture median was 3.00 (interquartile range [IQR] 3.00; N = 28) compared to non-agriculture 4.00 (IQR 2.00; N = 111), which was significant on Wilcoxon test (P = .015),so exposure score was included as a covariate in further analysis.

Stress symptoms
IES past score (first 7 days after event) was approximately normally distributed with overall mean 6.89, CI 6.39-7.39,minimum 0.00 and maximum 14.00 (N = 143).It had a negative Pearson correlation (−0.35,P < .001)with BRS score and a positive correlation (0.49, P < .001)with exposure score.Subjects marked a total of 4.3% of IES past symptoms as Don't Recall, and 0.8% of responses were missing.One hundred thirty-six subjects (95.1%) were able to recall Yes or No for at least 13 of the 15 symptoms, and only 1 did not recall more than half.Unadjusted, IES past score was significantly different by occupational group on t-test (P = .025;N Agriculture = 28, N Non-agriculture = 109).On multiple linear regression modeling, IES past score was significantly associated with sex (P < .001),exposure score (P < .001),and BRS score (P = .021).With these covariates, occupational group was not significant (P = .584).Age group, race or ethnicity, event type, dependents in home, and residence in or out of town also were not significant.Interactions were tested but rejected for insignificance.Assumptions of linearity, independence, normality, and equal variance were satisfied for the linear regression model.A Poisson regression model of symptom count was rejected for poor performance.

Symptom decline or persistence
Current stress symptoms.IES now score (past 30 days prior to taking survey) had an overall right skewed distribution with median 3.00 (IQR 5.00, N = 133).Wilcoxon testing was insignificant for different median IES now score by occupational group (P = .73;N Agriculture = 26, N Non-agriculture = 101).Subjects marked a total of 1.0% of IES now symptoms as Don't Recall, and 0.5% of responses were missing.There were 131 subjects (98.5%) able to recall Yes or No for at least 13 of 15 symptoms, and only 1 did not recall more than half.There were 112 (84.2%) who reported at least one disaster-associated symptom in the 30 days before taking the survey.On linear regression modeling with residual normality, occupational group was not significant (P = .081)when controlled for IES past, BRS, and exposure scores.
Subjects were also classified into IES past score rank groups above and below median to determine whether RR differed significantly between the groups with fewer initial symptoms and more initial symptoms.On Wilcoxon test, there was no evidence of a different RR between the median rank groups (P = .180).There were 17 participants who reported more symptoms in the month before the survey than in the week following the event, resulting in negative RRs.Of these participants, eight ranked in the lower half of IES past scores and nine were in the upper half.
On multiple linear regression, only BRS score was significantly associated with RR (P = .001),but BRS was not a good predictor of RR (R 2 = 0.073).Occupational group was not significant (P = .231)when controlled for BRS on regression or when controlled additionally for sex and exposure score (P = .370).Other covariates tested but excluded for insignificance were years since event, IES past score, age group, rural/urban, dependents in home, event type, exposure score, race or ethnicity, and sex.Normality of residuals was adequately satisfied for the model.

Posttraumatic growth
Of the subjects, 126 completed the PTGI-SF with overall mean score 26.56 (CI 24.18-28.09) of a possible 50 points, with approximately normal distribution.The agriculture group mean was 21.08 (CI 16.41-25.75)with minimum 0, maximum 36, and N = 24.The non-agriculture group mean was 27.67 (CI 25.38-29.96)with minimum 2, maximum 47, and N = 97.On unadjusted t-test, there is evidence that means are significantly different for agricultural and non-agricultural groups (P = .012;Table 2).Non-parametric Wilcoxon testing to confirm, due to distribution by subgroups, had P = .014,further evidence for significant difference in central measure between the groups.On multiple linear regression, PTGI score and IES past score were significantly positively associated (P < .001),and an occupational group by sex interaction was identified where women in agriculture had a significantly lower mean PTGI score (P = .024)(Table 3).Main effects were included in the model for occupational group and sex.Linearity, independence, normality, and equal variances were adequately satisfied on diagnostic plots.Model main effects of years since event, age group, IES now score, RR, exposure score, event type, and BRS score were rejected for insignificance.Occupation group interactions with these variables and with IES past score were also tested but insignificant.
When the analysis was re-run for only females by occupational group, the agricultural group (mean 19.4,CI 12.4-26.3,N = 14) scored significantly lower (P = .004)on PTGI-SF compared to non-agricultural (29.3, CI 26.9-31.7,N = 78) when controlled for IES past score (as indicated by the regression model) on analysis of covariance (ANCOVA).An unadjusted Wilcoxon test was conducted due to the small subgroup size and showed a significant difference (P = .005)in central measure (agriculture: median 23.5, IQR 22; non-agriculture: 32.0, IQR 15).Descriptive statistics to compare occupational group by sex PTGI scores are provided (Table 4).

Discussion
Stress and recovery around a disaster can be pictured as a cycle of baseline resilience, then the event, followed by stress symptoms, symptom decline and/or persistence, recovery toward baseline, and personal growth (Figure 2).Individuals may not experience all aspects, but all are considerations for comparing experiences.

Key results
We hypothesized that agricultural producers have different disaster stress and recovery experiences compared to their rural, non-agricultural counterparts across the stress and recovery cycle.We did not specify the direction of difference.In a systematic review, agricultural producers demonstrated higher occupational control and self-efficacy, which are potentially protective for mental health, 25 so they might be expected to score higher in resilience and recovery, and lower in posttraumatic stress, compared to non-agricultural residents.One researcher anticipated high resilience among farmers but found low resilience on analysis. 26,27A number of studies have shown that factors such as pesticide exposure, financial stressors, and weather concerns contribute to higher mental health risk in general for agricultural occupations than non-agricultural, but other studies have shown lower risk or no difference in rik. 28n the present study, no significant difference was identified between the agriculture and nonagriculture groups in resilience on BRS or posttraumatic stress symptom count on IES past.There was also no evidence of difference in recovery toward baseline on Recovery Ratio, which incorporated the IES now score into a proportional reduction in symptom count.The PTGI-SF qualitatively measured positive postevent growth and found an unadjusted significant difference between the agriculture and non-agriculture groups, as well as a sex by occupational group interaction on regression modeling.
Contrary to our hypothesis, this study found there is not strong evidence that agricultural producers have different overall experiences compared to rural, non-agricultural residents across the proposed disaster stress and recovery cycle.The agricultural participants did not exhibit a unique resilience or immunity to post-disaster stress effects.However, since women overall had significantly lower BRS score and higher IES past symptom count, and women in agriculture specifically reported lower posttraumatic growth, the results suggest that women in agriculture may be at risk of relatively lower resilience and lower posttraumatic growth compared to men in agriculture.Differences could exist due to baseline perceptions on qualitative scales, which still raise a question of why women in agriculture may report less confidence about recovery and lower personal growth after acute-onset disaster.A small sample size also could have influenced results.However, this finding is consistent with international studies that identified higher stress and lower resilience for agricultural women compared to men. 27Farm women may have greater stress due to performing multiple roles in farm work, off-farm work, and home responsibilities including childcare. 29Natural disasters disrupt social networks due to displacement and interrupted routines.Similar disruption was noted in the COVID-19 pandemic and may disproportionately affect farm women. 30Higher general stress may be associated with higher disaster stress and diminished recovery although this study was not designed to detect such a relationship.The stress and recovery experiences of women in agriculture merit further study and attention in disaster preparedness and recovery planning.

Implications for disaster preparedness, response, and recovery
While we exceeded the total sample estimate with 159 in the data analysis set, the allocation ratio was closer to 1:4 than the planned 1:1.For this reason, we completed a post hoc power calculation based on the actual group sizes at each comparison.Post hoc power for a medium effect size d = 0.5 ranged from 59% to 71%, and for a large effect size d = 0.8 from 94% to 98%.Though acknowledging concerns about the value of post hoc power analysis, 31 we believe the sample size was adequate for identifying a large difference between agricultural and non-agricultural groups.It is unlikely that small-to-medium effects would require differential policies or programs, so this study provides evidence for including agricultural residents in general community response and recovery plans. 32As noted, women in agriculture may benefit from targeted strategies.
In this study, some rural residents reported posttraumatic-type symptoms persisting up to 8 years beyond acute-onset natural disaster events.Community-level actions mindful of disasterrelated emotional health could be beneficial over an extended period of time.In another study, 2 years post-flood, rural community members described long-term stresses and how some response activities supported or hindered resilience. 32[35]

Strengths and limitations
A strength of this study was the comprehensive approach to evaluating a cycle of disaster stress and recovery experiences in rural populations with intentional efforts to include agricultural residents.Using a combination of existing scales, data represented a story of community experiences over time although it was collected at a single time point.Another strength was the presence of committed local individuals to recruit participants in the targeted communities, an important factor in rural culture.
Limitations centered around study and survey design.The study used a voluntary convenience sample prone to selection bias, and rural disasteraffected populations, especially agricultural, are challenging to reach and may be uncertain about participating in outside research.Local collaborators indicated that internet access and use is unreliable for survey distribution in this population.Paper surveys were mailed based on affected ZIP codes and local knowledge of affected residents.In the analysis data set, females were over-represented, particularly in the non-agricultural group; however, sex was controlled in statistical modeling.Multiple household members could have completed the survey, potentially introducing correlation in the data although family members may also experience different stress and recovery patterns. 36Finally, the study included severe acute-onset natural disasters -primarily tornado and flood -in the South Central and Midwestern U.S. Results may not generalize to disasters of other types or intensities, or to diverse geographic communities with unique culture and resources.
The RNDSR survey has inherent limitations.Exposure questions, BRS, and PTGI-SF have qualitative components subject to variability.Whether a participant's own definition of "directly affected" or "financial hardship," or a personality unlikely to "strongly agree" or "strongly disagree," quantitative analysis of qualitative data requires caution.Trends and patterns bear more weight than specific numeric values or individual scores.
The use of the Revised IES was modified to ask about symptoms occurring years before during the week following a disaster event and is subject to recall bias.To address this, we added the option Don't Recall; however, only 4.3% of response items were marked as Don't Recall, and only 0.8% were missing.There is evidence of strong and longlasting recall around disaster events 37 to support delayed inquiry.The Revised IES is also limited to questions about intrusion (re-experiencing) and avoidance posttraumatic stress symptoms without addressing hyperarousal posttraumatic symptoms or any other mental health indicators such as depression, anxiety, or substance abuse.Frequency, intensity, and life disruption of symptoms were not accounted for, only presence or absence of symptoms.The findings in this study do not preclude the possibility of additional or disparate severity of emotional or mental health challenges by occupational subgroup.
This study broadly targeted affected individuals through both agricultural and general community sources, but the survey did not distinguish between farmworkers and farm owner/operators, and the study was not designed to capture experiences of a mobile population.It is likely that the agricultural participants were primarily owners.Migrant farmworkers, in particular, are recognized as a vulnerable population for whom stress factors and mental health risks are not well understood. 38,39Future study to evaluate migrant farmworkers' natural disaster stress and recovery is warranted with attention to mobile lifestyle and language preferences.

Conclusion
While the comparison between agricultural and rural, non-agricultural groups did not yield statistically significant differences overall, the RNDSR study has implications for disaster management and future research.Agricultural populations should be included alongside non-agricultural counterparts in disaster mental and emotional health supports, and women in agriculture may require specific attention.The long-term persistence of posttraumatic stress symptoms for many individuals should also be a factor in community level disaster recovery plans.Additional efforts should be made to study effects in agricultural populations where a larger sample size can participate and where further comparisons can be made between men and women.The RNDSR survey is easily adaptable to additional disaster settings for future studies.

Figure 1 .
Figure 1.Sample size flow for rural and agricultural natural disaster stress and recovery study.

Figure 2 .
Figure 2. Disaster stress and recovery cycle as proposed by authors.

Table 1 .
Demographic summary of participants in data analysis set of the rural and agricultural natural disaster stress and recovery study (N = 159).
cNine of the missing event-related values were participants who had not experienced a disaster and completed only the BRS.

Table 2 .
Unadjusted t-test for PTGI score by occupational group (agricultural vs. nonagricultural).

Table 3 .
Adjusted linear regression model for PTGI score.

Table 4 .
Descriptive statistics for occupational groups by sex.Effect of sex differs by occupational group.