Displacement, county social cohesion and depression after a large-scale traumatic event

Abstract

Background

Depression is a common and potentially debilitating consequence of traumatic events. Mass traumatic events cause wide-ranging disruptions to community characteristics, influencing the population risk of depression. In the aftermath of such events, population displacement is common. Stressors associated with displacement may increase risk of depression directly. Indirectly, persons who are displaced may experience erosion in social cohesion, further exacerbating their risk for depression.

Methods

Using data from a population-based cross-sectional survey of adults living in the 23 southernmost counties of Mississippi (N = 708), we modeled the independent and joint relations of displacement and county-level social cohesion with depression 18–24 months after Hurricane Katrina.

Results

After adjustment for individual- and county-level sociodemographic characteristics and county-level hurricane exposure, joint exposure to both displacement and low social cohesion was associated with substantially higher log-odds of depression (b = 1.34 [0.86–1.83]). Associations were much weaker for exposure only to low social cohesion (b = 0.28 [−0.35–0.90]) or only to displacement (b = 0.04 [−0.80– 0.88]). The associations were robust to additional adjustment for individually perceived social cohesion and social support.

Conclusion

Addressing the multiple, simultaneous disruptions that are a hallmark of mass traumatic events is important to identify vulnerable populations and understand the psychological ramifications of these events.

INTRODUCTION

Large-scale traumatic events cause multiple, simultaneous disruptions to communities’ physical and social environments. Current understanding of how this broad range of disruptions affects mental health is limited. Behavioral health studies often address the impacts of single exposures, failing to take into account the simultaneous influence of other changes to the environment. We examined the joint influence of post-event displacement and community social cohesion, two factors that have been individually related to mental health, on the risk of depression in a population affected by a natural disaster. Depression is a common and potentially debilitating consequence of disasters [1–3].

Nearly 15 million people were displaced because of natural disasters in 2011, including over 30,000 in the United States, and an additional 3.5 million were newly displaced within their countries because of violence and conflict [4,5]. Displaced persons may experience higher risk for depression than the nondisplaced because in addition to the trauma of the event itself, they may experience a host of additional stressors that unfold and persist afterwards [6–8]. In addition to material and administrative hardships such as decreased access to medical and social services, worse living conditions, repair or replacement of damaged possessions, application processes for governmental aid, and difficulty finding employment, the loss of social networks, social supports, and normal routines can add to the psychological burden of displacement. Extant research suggests that postdisaster stressors may be an important cause of depression in disaster-affected populations [9,10] and, more generally, has related stressful life events with depression [11,12].

Within the framework of the socioecological model of health—in which individuals are embedded in, and their health affected by, nested levels of social and physical environments [13]— mental health following a large-scale traumatic event may vary depending not only on attributes of affected individuals but also on attributes of their communities [14–17]. Community social cohesion may be one such attribute. Social cohesion describes the attachments between individuals and their

communities; these attachments may be demonstrated in feelings of belonging, willingness and ability to cooperate, and shared values among community members [18]. A small body of work has related residence in communities characterized by high social cohesion with a lower risk of mental health problems, including depression, after taking into account individual-level characteristics [19–21].

After a traumatic event, socially cohesive communities may be better able to respond quickly and effectively by organizing and carrying out relief efforts, gaining aid from outside organizations, assessing damage and identifying community members most in need, mitigating secondary problems such as looting and illness, disseminating information, and organizing rebuilding projects [16,17]. Social cohesion may also provide social supports that buffer community members’ experiences of the stressors they do encounter by causing residents to interpret the stressors as less stressful, experience less severe physiological responses to the stressors, or be less likely to respond to the stressors in potentially harmful ways (such as substance abuse) [22,16].

These material and psychosocial benefits of social cohesion may be of particular importance for displaced persons burdened with numerous stressors in the aftermath of the event. In addition, it is plausible that persons who are displaced experience erosion in social cohesion, further exacerbating their risk for depression. Persons jointly exposed to both displacement and low community social cohesion may therefore be a particularly vulnerable population with high risk of depression.

We investigated the independent and joint relations of displacement and county-level social cohesion with depression among residents of the state of Mississippi 18–24 months after Hurricane Katrina. We hypothesized that displaced residents would be at higher risk for depression than nondisplaced residents, and that residents of counties with low social cohesion would be at higher risk than residents of counties with high social cohesion. We also hypothesized that residents who had both been displaced and resided in counties with low social cohesion would experience a higher risk of depression than would be expected from the independent effects of the two exposures.

METHODS

Study Population

Data came from the Mississippi Community Survey (MCS), a population-based survey of adults 18 years and older living in the 23 southernmost counties of Mississippi prior to Hurricane Katrina. Figure 1 shows the study area. An in-depth discussion of the sampling methods is available elsewhere [23]. Briefly, the sampling frame was created through the enumeration of all addresses existing prior to Katrina in randomly selected census-based segments of areas severely affected by the storm. Extensive tracking using multiple methods was conducted to locate residents who had been living at selected addresses prior to the storm. Random-digit dialing was used in less-damaged areas where there was less hurricane-related resident relocation. One participant was randomly selected from the eligible adults in each sampled household. Interviews were conducted in person and by phone between February and July 2007. Interviews lasted 37 minutes on average and collected information about sociodemographic characteristics, hurricane-related experiences, individual and community support and coping experiences, traumatic events other than Katrina, and mental health symptoms. The study received Institutional Review Board approval from the University of Michigan and respondents gave oral consent to participate.

Figure 1.

Map of Mississippi counties included in the study area, by tertile of social cohesion. Black, medium gray, and light gray areas show sample counties with high, medium, and low social cohesion, respectively. White area shows surrounding counties in the states of Mississippi, Louisiana, and Alabama.

The community-level units of analysis were counties; each participant’s county of residence was identified using the address he/she provided during the interview. The 82 counties comprising the state of Mississippi are well-defined and administratively meaningful units. Each has its own elected board of supervisors, as well as departments providing social and administrative services. We felt that counties’ social and administrative relevance to residents made them the most appropriate choice of unit of analysis in this context despite the fact that they are relatively large and heterogeneous. Smaller units would not reflect the same policy and service differences and, in the case of census-defined areas, might not be geographically meaningful to residents. Several previous studies have linked county-level attributes, including characteristics of the social environment, with health [24–27].

This analysis was limited to self-reported non-Hispanic black or white respondents who resided in the same county both before and after Hurricane Katrina. The small number of respondents from other racial or ethnic groups (52 [7%] respondents, including 16 [2%] who reported being Hispanic) would have precluded drawing meaningful conclusions about members of these groups, particularly because the social environment may affect health differently for minority groups in areas such as our study area, in which the non-Hispanic black and white populations make up the overwhelming majority (93–99%). Similarly, forty-six non-Hispanic black or white MCS respondents were excluded because they moved to a different county after Katrina. We would expect the role of social cohesion to differ for these respondents because of their lack of a residential history in the counties where they lived at the time of the interview. These excluded respondents were more likely to have experienced stressors after Katrina related to the hurricane (90% vs. 76% of the included respondents), including displacement (75% vs. 26% of the included respondents). They also tended to be younger and were more likely to be married, although their low number makes meaningful comparisons difficult.

Measures

Displacement was assessed with the question, “Did you move from the place you were living because of Hurricane Katrina?” As mentioned above, at the time of the interview all sample participants resided in the same county as before Katrina. We included a separate measure of whether the respondent moved (within the county) after Katrina as a potential confounder.

Social cohesion was measured using a modified version of the 5-item scale developed by Sampson et al.; the scale has been validated against other measures of community integration and participation [28,29]. Respondents were asked to respond on a 5-point scale how much they agreed that people in their neighborhood after Hurricane Katrina 1) were close-knit or unified, 2) were willing to help their neighbors, 3) generally got along with each other, 4) did not share the same values, and 5) could be trusted. Each respondent’s responses to the 5 items were averaged (item 4 was reverse coded) to give a measure of individually perceived social cohesion (Cronbach’s α = .75). The individual social cohesion values for all respondents within each county were averaged to give an estimate of county-level social cohesion, a method analogous to methods used in previous research [28,29]. One consideration is that while our community units of analysis were counties, the questions referred to respondents’ neighborhoods. Our measure of county social cohesion therefore represents an aggregate measure of neighborhood social cohesion: a county with high social cohesion is interpretable as a county whose constituent neighborhoods have high levels of social cohesion. We categorized the social cohesion measure into tertiles at the county level; because the number of participants per county varied, the number of individual participants in each tertile of county social cohesion differs. Figure 1 shows the geographic distribution of counties by social cohesion level.

Depression was measured using a variation of the Patient Health Questionnaire-9 (PHQ-9) [30]. The PHQ-9, whose 9 items correspond to the Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV) symptoms of major depression, has demonstrated good psychometric properties and construct validity [31]. Respondents described how often they were bothered by each symptom in the time since Hurricane Katrina using a 4-point scale ranging from “not at all” to “nearly every day.” The items were scored 0–3 and then summed to create a score ranging 0–27 (Cronbach’s α = .89). In keeping with validity studies and other previous research, a cut-off score of 10, along with the respondent’s report that the symptoms were experienced within the same 2-week period, was used as the case definition for probable depression since Hurricane Katrina in this analysis [30,31]. In validation studies, using a cut-off of 10 the sensitivity and specificity of the PHQ-9 relative to clinician diagnosis of depression were both 88% [30]. As a secondary outcome, we also examined past-month depression, defined as meeting the case definition for depression and reporting having experienced the symptoms within the past month.

Social support was assessed using the 6-item Crisis Support Scale developed by Joseph et al. [32]. Respondents used a 7-point scale to describe how often they received various informal supports (such as sympathy, willingness to listen, and practical help) from others during the 2-month period after Katrina. Each individual’s responses were summed to give a score ranging 6–42 (Cronbach’s α = .77). Scores were categorized into tertiles for this analysis.

Potential individual-level confounders included in the analysis were age, gender, race, marital status, educational attainment, and pre-Katrina household annual income. We also included information about four lifetime posttraumatic stress disorder (PTSD) symptoms related to an event other than Katrina: dwelling on or having nightmares about the event, avoiding thinking about the event or situations that reminded the respondent of the event, feeling constantly on guard or easily startled, and feeling numb or detached [33]. Although depression and PTSD are distinct illnesses, with distinct but overlapping causes, the relatively high rate of comorbidity of these conditions after disasters, along with the monotonic relation between lifetime PTSD symptoms and depression in our sample, suggested that the use of this measure was appropriate in this context to address potential confounding of our results by participants’ prior mental health [1,10]. A census-based measure of the percent of residents aged 25 or over with a college degree was also included as a measure of county socioeconomic status (results were nearly identical when this measure was replaced with county-level median household income). The percent of occupied housing units with major or severe damage from Katrina, as classified by the Federal Emergency Management Agency (FEMA), was included as a measure of county-level hurricane exposure [34].

Analysis

All analyses were conducted with SAS software (version 9.2) [35] using SURVEY procedures to account for the study sample design and possible correlations between residents of the same counties induced by the multilevel nature of the data [36]. Analyses alternatively using a general estimating equations (GEE) approach to account for within-county correlations produced nearly identical results [37]. Analyses incorporated weights to account for household and within-household individual sampling probabilities and additionally including a poststratification factor to reflect the age, gender, race/ethnicity, and educational attainment make-up of the study area according to the 2000 U.S. Census. Multiple imputation with 25 imputations was used to impute missing data while accounting for the uncertainty inherent in doing so [38]. At least one variable was imputed for 86 participants (12%); for 81 of these, the only missing information was the measure of household income.

We used unadjusted multinomial regression to examine bivariable associations between the covariates and county social cohesion, and unadjusted logistic regression to examine associations between the covariates and odds of both displacement and depression. We then used logistic regression to examine crude and confounder-adjusted associations of both displacement and county social cohesion with odds of depression. We included interaction (product) terms in our models to investigate joint effects of displacement and social cohesion. Finally, we tested the robustness of our results to additional adjustment for individually perceived social cohesion and social support measures. To the extent that individually perceived social cohesion and social support are influenced by county social cohesion, adjusting for these measures may result in an underestimate of the total effect of county social cohesion on depression. However, given that the county social cohesion variable used in this analysis was constructed by aggregating individual responses, it may be appropriate to view individually perceived social cohesion and social support as confounders in this case. We present results both with and without adjustment for these measures.

RESULTS

Participation rates in the study were comparable to or better than those documented in other population-based studies carried out under similarly difficult conditions: interviews were completed by 50.3% of all eligible sampled households, and only 9.4% of households that were sampled and successfully contacted refused to participate [23,39]. After application of the weights, the study population was sociodemographically representative of the 2000 U.S. Census population in the study area [23]. Out of 810 total participants in the study, 708 met inclusion criteria for this analysis. Sample participants were aged 18–91; after weighting, 52% of the sample was female, 25% was black, 57% was married, 45% had schooling beyond a high school diploma, 50% had an annual household income of at least $40,000, and 26% had been displaced after Hurricane Katrina (Table 1). The number of participants in each sample county ranged from 2–168 with a median of 20. The range of county-level social cohesion was relatively limited (range 2.3–3.6, with a possible range 0–4). Mean social cohesion scores for each tertile were 2.62, 3.01, and 3.39 for low-, medium-, and high-cohesion counties, respectively (F = 50.2; p < .0001). County socioeconomic status and degree of Katrina exposure were more varied (Table 1).

Table 1. Sociodemographic characteristics of sample population and sample area, Mississippi Community Study, 2007^a.

Frequencies and weighted percents of the sample population by category of individual sociodemographic characteristics, as well as medians and ranges of the sample counties by category of county-level characteristics.

	Individual		County
Characteristic	N	Weighted %b	Median	Range
Gender
Female	443	52
Male	265	48
Age
18–34	110	28
35–64	411	53
65–91	187	19
Race
Black	148	25
White	560	75
Marital statusc
Married/living with partner	390	57
Divorced/separated/widowed	226	21
Never married	90	22
Missing	2	<1
Educational attainmentc
Less than high school	99	24
High school/GED	233	32
Some college	199	31
College graduate	176	13
Missing	1	<1
Annual household income before Katrinac
<$20,000	151	27
$20,000-$39,999	193	31
$40,000-$59,999	111	15
≥$60,000	170	27
Missing	83	12
Displacement after Katrina
No	499	74
Yes	208	26
Moved (within county) after Katrina
No	652	92
Yes	56	8
Lifetime PTSD symptoms
0	328	48
1–2	220	28
3–4	160	24
Individual social cohesion score
0–2 (Low)	82	16
2.1–3	189	24
3.1–4 (High)	437	60
Missing	0	0
Social support tertilec
Low	189	28
Medium	234	32
High	284	39
Missing	1	<1
Social cohesion (possible range 0–4; higher score indicates higher social cohesion)			2.9	2.3–3.6
% population aged 25+ with at least a BA			7.5	4.7–16.4
Median annual household income 2000			$2.7K	$1.9K–$3.9K
% occupied housing units with major or severe damage from Hurricane Katrina			3.4	0.2–69.8

^aSample limited to respondents who reported being of non-Hispanic black or white race and who did not change their county of residence after Hurricane Katrina (N = 708). Sample area includes 23 counties.

^bMay not add to 100% because of rounding.

Participants who had been displaced had lower individual social cohesion scores, were more likely to live in counties severely affected by Katrina, and were more likely to meet the case definition for depression (Table 2). Displacement status also varied by individual socioeconomic status and having experienced PTSD symptoms from an event other than Katrina, although not monotonically. Residence in a county with low social cohesion was associated with younger age, lower education, and having experienced PTSD symptoms from an event other than Katrina (Table 2). It was also strongly associated with being displaced or moving after Hurricane Katrina. Higher county social cohesion was strongly associated with higher individually rated social cohesion, which is to be expected not only theoretically but because of the construction of the county measure. County socioeconomic status and hurricane exposure were not related to county social cohesion (p = 0.87 and p = 0.28, respectively) (Table 2). Table 3 shows coefficients (labeled b) from unadjusted and adjusted logistic regression models of depression. Respondents were more likely to meet the case definition for depression if they were younger, were of black race, had a lower education level or income, or reported low social support or individual social cohesion (Table 3, Model 1).

Table 2. Sample characteristics by displacement and tertile of county social cohesion, Mississippi Community Study, 2007^a.

The top section shows the joint distribution (frequency and weighted percent) of the sample population by categories of individual characteristics and both displacement status and social cohesion tertile. The bottom section shows the joint distribution (median and range) of the sample population by county-level characteristics and both displacement status and social cohesion tertile.

	Displacementb						Social Cohesionb
	No	Yes	No	Yes	X2	p	High	Med	Low	High	Med	Low	X2	p
Individual characteristics	N	N	%	%			N	N	N	%	%	%
Total	499	209	74	26			131	209	368	23	28	49
Gender					1.8	0.19							1.0	0.61
Female	312	131	77	23			88	129	226	25	26	49
Male	187	78	70	30			43	80	142	22	29	49
Age					1.9	0.39							8.4	0.08
18–34	79	31	72	28			13	31	66	16	27	57
35–64	289	122	72	28			77	120	214	25	27	48
65–91	131	56	79	21			41	58	88	30	30	40
Race					1.3	0.26							0.6	0.73
Black	102	46	68	32			28	52	68	20	28	52
White	397	163	75	25			103	157	300	25	27	48
Marital status					1.8	0.64							6.2	0.19
Married/living with partner	280	110	73	27			77	116	197	27	27	46
Divorced/separated/widowed	151	77	69	31			43	71	114	21	30	49
Never married	68	22	80	20			11	22	57	17	25	57
Educational attainment					25.7	<0.001							14.2	0.03
Less than high school	70	29	66	34			19	24	56	20	19	61
High school/GED	177	57	81	19			41	69	124	23	30	47
Some college	143	56	76	24			34	57	108	24	28	48
College graduate	109	67	63	37			37	59	80	29	36	35
Annual household income before Katrina					11.5	0.01							7.2	0.30
<$20,000	106	47	67	33			20	34	99	15	21	65
$20,000-$39,999	167	65	77	23			41	70	121	24	29	47
$40,000-$59,999	102	49	67	34			30	45	76	22	29	48
≥$60,000	124	48	80	20			40	60	72	32	31	37
Experienced a Katrina-related traumatic event					5.9	0.02							2.4	0.30
No	374	126	77	22			105	162	233	25	30	45
Yes	125	83	64	36			26	47	135	19	22	59
Displacement after Katrina													24.0	<0.001
No	--	--	--	--		--	107	164	228	28	29	43
Yes	--	--	--	--		--	24	45	140	11	23	65
Moved (within county) after Katrina					3.6	0.06							3.1	0.21
No	475	177	76	24			123	197	332	23	29	48
Yes	24	32	53	47			8	12	36	24	16	60
Number of lifetime PTSD symptoms					28.1	<0.001							19.3	0.01
0	246	82	81	19			74	96	158	29	29	42
1	73	37	71	29			14	35	61	16	25	59
2	74	36	62	38			16	33	61	20	23	57
3	59	30	66	34			13	27	49	14	39	47
4	47	24	68	32			14	18	39	21	19	60
Individual social cohesion score					14.4	<0.001							144.2	<0.001
0–1.9 (Low)	50	31	59	41			5	20	57	7	20	73
2–2.9	129	60	72	28			21	65	103	13	31	56
3–4 (High)	319	118	78	22			105	124	208	32	28	40
Social support tertile					1.8	0.41							3.0	0.56
Low	140	49	75	25			36	57	96	21	27	52
Medium	155	80	69	31			36	67	132	22	24	54
High	204	80	76	24			59	85	140	26	31	43
Depression					17.3	<0.001							10.8	0.005
Yes	68	50	57	43			14	26	78	15	18	67
No	431	159	78	22			117	183	290	26	30	44
County characteristics	Median	Median	Range	Range			Median	Median	Median	Range	Range	Range
Median household annual income ($10,000 units)c	2.9	3.5	1.9- 3.9	1.9- 3.9	2.8	0.09	2.6	2.7	2.9	2.5- 3.8	2.3- 3.9	1.9- 3.6	0.2	0.89
County % population aged 25+ with at least a BAc	8.9	10.4	4.7- 16.4	4.7- 16.4	1.0	0.32	7.8	6.7	7.6	6.2- 16.4	4.7- 13.1	5.8- 11.6	0.3	0.87
County % occupied housing with major/severe damage from Katrina	4.1	34.2	0.20- 69.8	0.40- 69.8	9.4	0.002	1.0	4.0	5.4	0.2- 5.2	0.3- 34.2	1.9- 3.6	2.5	0.28

^aSample limited to respondents who reported being of non-Hispanic black or white race an d who did not change their county of residence after Hurricane Katrina. Information incorporates imputed values and percents are weighted.

^bFor displacement the population is individual respondents and p-values are from unadjusted logistic regression.For social cohesion, the population is individual respondents (N = 708) for individual characteristics and counties (N = 23) for the county characteristics; p-values are from unadjusted multinomial logistic regression.

^cData from 2000 U.S. Census.

Table 3. Coefficients from logistic regression models of depression since Hurricane Katrina, Mississippi Community Study, 2007^a.

Coefficients and 95% confidence intervals from logistic regression models of depression since Hurricane Katrina. Coefficients in Model 1 are not adjusted for other variables. Model 2 adjusted for displacement, low county social cohesion, Black race, sex, age, educational attainment, number of lifetime PTSD symptoms, county percent population aged 25+ with at least a bachelor’s degree, and county percent occupied housing units with major or severe damage from Hurricane Katrina. Model 3 = Model 2 + displacement x low county social cohesion interaction. Model 4 = Model 3 + individual perceived social cohesion score. Model 5 = Model 3 + social support tertile. Model 6 = Model 3 + individual perceived social cohesion score and social support tertile.

	Unadjusted		Adjusted
	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
−2 Log Likelihoodb			588.91		584.63		584.41		578.47		578.30
Covariate	b	95% CI	b	95% CI	b	95% CI	b	95% CI	b	95% CI	b	95% CI
Displaced after Katrina	0.97	(0.51, 1.43)***	0.72	(0.15, 1.30)*	0.04	(−0.80, 0.88)	0.03	(−0.79, 0.85)	0.00	(−0.85, 0.85)	0.01	(−0.84, 0.86)
Low social cohesion	0.94	(0.38, 1.51)**	0.58	(0.04, 1.12)*	0.28	(−0.35, 0.90)	0.25	(−0.43, 0.93)	0.23	(−0.39, 0.86)	0.26	(−0.42, 0.93)
Low social cohesion x displaced					1.02	(0.05, 2.00)*	1.03	(0.06, 1.99)*	1.11	(0.15, 2.08)*	1.11	(0.14, 2.09)*
Black race (vs. White)	0.98	(0.01, 1.95)*	0.62	(−0.26, 1.50)	0.71	(−0.18, 1.61)	0.69	(−0.22, 1.61)	0.67	(−0.27, 1.61)	0.69	(−0.24, 1.62)
Female gender	0.12	(−0.24, 0.49)	0.08	(−0.32, 0.48)	0.11	(−0.30, 0.52)	0.11	(−0.31, 0.54)	0.10	(−0.33, 0.53)	0.10	(−0.34, 0.54)
Age (per 5 years)	−0.09	(−0.17, 0.00)*	−0.06	(−0.15, 0.02)	−0.06	(−0.15, 0.02)	−0.06	(−0.15, 0.02)	−0.06	(−0.15, 0.02)	−0.07	(−0.15, 0.02)
Educational attainment (vs. <high school)
High school/GED	−1.05	(−1.85, −0.24)*	−0.68	(−1.53, 0.17)	−0.62	(−1.50, 0.27)	−0.59	(−1.38, 0.20)	−0.60	(−1.51, 0.31)	−0.62	(−1.43, 0.18)
Some college	−1.23	(−2.14, −0.32)**	−0.99	(−1.80, −0.18)*	−0.95	(−1.80, −0.11)*	−0.92	(−1.69, −0.15)*	−0.92	(−1.76, −0.07)*	−0.95	(−1.72, −0.17)*
College graduate	−1.89	(−2.77, −1.02)***	−1.54	(−2.34, −0.75)***	−1.40	(−2.23, −0.57)**	−1.35	(−2.09, −0.62)***	−1.31	(−2.17, −0.44)**	−1.34	(−2.10, −0.58)***
Number of lifetime PTSD symptoms	0.55	(0.34, 0.75)***	0.49	(0.24, 0.73)***	0.49	(0.25, 0.73)***	0.49	(0.26, 0.72)***	0.45	(0.21, 0.70)***	0.46	(0.22, 0.70)***
County % population aged 25+ with a BAc	−0.11	(−0.26, 0.04)	−0.05	(−0.19, 0.10)	−0.05	(−0.19, 0.08)	−0.05	(−0.19, 0.09)	−0.05	(−0.19, 0.09)	−0.05	(−0.19, 0.09)
County % occupied housing units with major/severe damage from Katrina (per 5%)	−0.01	(−0.11, 0.09)	−0.01	(−0.11, 0.08)	−0.01	(−0.09, 0.08)	−0.01	(−0.09, 0.08)	−0.01	(−0.10, 0.08)	−0.01	(−0.10, 0.08)
Individual social cohesion scored	−0.67	(−1.09, −0.22)**					−0.06	(−0.42, 0.31)			0.05	(−0.34, 0.44)
Social support (vs. T3: High)
T2: Medium	0.89	(0.12, 1.66)*							0.27	(−0.57, 1.11)	0.28	(−0.55, 1.11)
T1: Low	1.28	(0.41, 2.14)**							0.66	(−0.30, 1.63)	0.71	(−0.27, 1.68)

^aSample limited to respondents who reported being of non-Hispanic black or white race and who did not change their county of residence after Hurricane Katrina. N=708; weighted N=1533.

^bAverage value over 25 imputations; −2 log likelihood for intercept-only model = 743.42; −2 log likelihood for model including only social cohesion, displacement, and their interaction = 698.30.

^cData from 2000 U.S. Census.

^dRange 0–4. Higher represents higher social cohesion.

^*p < 0.10;

^**p < 0.01;

^***p < 0.001

In models including both medium and high county social cohesion, estimated effects of medium social cohesion did not differ substantively from the null. Therefore, in the final models presented here we combined these two groups into one reference group; doing this did not affect the point estimates for low social cohesion. In unadjusted models, respondents were more likely to meet the case definition for depression since Hurricane Katrina if they had been displaced (Table 3, Model 1; b [coefficient] = 0.94 [95% confidence interval] 0.51–1.43) or lived in a county with low social cohesion (b = 0.94 [0.38– 1.51]). After adjustment for individual- and county-level confounders, the estimates were attenuated somewhat (Model 2; b = 0.72 [0.15, 1.30] for displacement, b = 0.58 [0.04–1.12] for low social cohesion). As we hypothesized, there was evidence of interaction between displacement status and social cohesion (Model 3; interaction p = 0.04). Participants exposed only to low social cohesion experienced moderately elevated log-odds of depression, although the confidence interval spanned the null (b = 0.28 [−0.35–0.90]) while exposure only to displacement was not independently associated with appreciably different log-odds of depression (b = 0.04 [−0.80–0.88]). However, joint exposure to both displacement and low social cohesion was associated with substantially higher log-odds of depression than exposure to neither condition (calculated from Model 3 coefficients; b = 1.34 [0.86–1.83]).

The final models do not include marital status or whether the participant moved (within the county) after Katrina; these variables were not associated with either the exposure or the outcome in the bivariable analysis and their exclusion from the multivariable models did not affect the results for social cohesion. Similarly, because results were nearly identical when models included individual education, individual income, or both measures, we present models including only individual education. Figure 2(a) shows predicted probabilities of depression from Model 3, by social cohesion and displacement status.

Figure 2.

Predicted probabilities of (a) depression since Hurricane Katrina and (b) past-month depression, by county social cohesion and displacement status. Probabilities calculated from Table 3, Model 3 and Table 4, Model 3, respectively. Models adjusted for gender, age, race, education, lifetime PTSD symptoms, county education level, and county exposure to Hurricane Katrina. Probabilities calculated for a White, 54-year-old (population mean) male with a high school degree, no lifetime PTSD symptoms, mean county education level, and median county Katrina exposure.

Table 3 also shows results from models additionally adjusted for individually perceived social cohesion (Model 4), social support (Model 5), and both individually perceived social cohesion and social support (Model 6). These additional adjustments did not appreciably affect the estimates for displacement, county social cohesion, or their statistical interaction.

Ninety-one participants (weighted 15%) reported experiencing past-month depression. Results from models using past-month depression as the outcome were consistent with those for any depression since Katrina, although less precise (Table 4 and Figure 2[b]). Displacement and low social cohesion were both independently associated with higher odds of past-month depression after adjustment (Model 2; b = 1.26 [0.68–1.84] and b = 0.66 [0.09–1.24], respectively); compared to the models of depression since Hurricane Katrina, the main-effect point estimate for displacement was larger in magnitude while the estimate for low social cohesion was smaller. Just as with the models of any depression, there was evidence of interaction (interaction p = 0.02) such that joint exposure to both displacement and low social cohesion was related to substantially higher odds of past-month depression (calculated from Model 3 coefficients; b = 1.83 [1.37–2.30]) and main-effect estimates for both exposures were minimal in the interaction model (b = 0.17 [−0.99–1.33] and b = 0.12 [−0.55–0.78], respectively). Also consistent with models of depression since Katrina, the estimates were not substantially changed by additional adjustment for individually perceived social cohesion and social support (Models 4–6).

Table 4. Coefficients from logistic regression models of past-month depression, Mississippi Community Study, 2007^a.

Coefficients and 95% confidence intervals from logistic regression models of past-month depression. Coefficients in Model 1 are not adjusted for other variables. Model 2 adjusted for displacement, low county social cohesion, Black race, sex, age, educational attainment, number of lifetime PTSD symptoms, county percent population aged 25+ with at least a bachelor’s degree, and county percent occupied housing units with major or severe damage from Hurricane Katrina. Model 3 = Model 2 + displacement x low county social cohesion interaction. Model 4 = Model 3 + individual perceived social cohesion score. Model 5 = Model 3 + social support tertile. Model 6 = Model 3 + individual perceived social cohesion score and social support tertile.

	Unadjusted		Adjusted
	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
−2 Log Likelihoodb			471.67		464.13		462.55		447.92		447.92
Covariate	b	95% CI	b	95% CI	b	95% CI	b	95% CI	b	95% CI	b	95% CI
Displaced after Katrina	1.40	(0.85, 1.95)***	1.26	(0.68, 1.84)***	0.17	(−0.99, 1.33)	0.14	(−1.00, 1.27)	0.13	(−1.11, 1.36)	0.13	(−1.09, 1.35)
Low social cohesion	1.18	(0.61, 1.76)***	0.66	(0.09, 1.24)*	0.12	(−0.55, 0.78)	0.04	(−0.70, 0.77)	0.01	(−0.63, 0.65)	0.01	(−0.69, 0.72)
Low social cohesion x displaced					1.55	(0.20, 2.90)*	1.57	(0.25, 2.90)*	1.71	(0.30, 3.12)*	1.71	(0.30, 3.12)*
Black race (vs. White)	0.50	(−0.60, 1.60)	−0.14	(−0.92, 0.64)	0.03	(−0.75, 0.82)	−0.03	(−0.83, 0.77)	−0.09	(−0.94, 0.76)	−0.09	(−0.93, 0.76)
Female gender	0.28	(−0.27, 0.83)	0.42	(−0.02, 0.86)*	0.48	(0.07, 0.89)*	0.50	(0.05, 0.95)*	0.50	(0.07, 0.93)*	0.50	(0.05, 0.95)*
Age (per 5 years)	−0.11	(−0.22, 0.01)*	−0.09	(−0.20, 0.02)	−0.09	(−0.20, 0.02)	−0.09	(−0.20, 0.03)	−0.09	(−0.20, 0.02)	−0.09	(−0.20, 0.02)
Educational attainment (vs. < high school)
High school/GED	−1.15	(−2.05, −0.24)*	−0.72	(−1.63, 0.19)	−0.64	(−1.59, 0.31)	−0.54	(−1.48, 0.39)	−0.58	(−1.60, 0.44)	−0.59	(−1.56, 0.39)
Some college	−1.09	(−2.10, −0.09)*	−0.83	(−1.73, 0.07)*	−0.79	(−1.76, 0.18)	−0.68	(−1.59, 0.24)	−0.64	(−1.64, 0.35)	−0.65	(−1.60, 0.30)
College graduate	−1.59	(−2.49, −0.69)***	−1.31	(−2.05, −0.56)***	−1.08	(−2.01, −0.16)*	−0.94	(−1.75, −0.12)*	−0.86	(−1.94, 0.21)	−0.87	(−1.84, 0.10)*
Number of lifetime PTSD symptoms	0.60	(0.39, 0.82)***	0.53	(0.27, 0.78)***	0.54	(0.29, 0.79)***	0.52	(0.28, 0.76)***	0.48	(0.22, 0.73)***	0.48	(0.23, 0.72)***
County % population aged 25+ with at   least a BAc	−0.10	(−0.26, 0.07)	−0.13	(−0.23, −0.02)*	−0.13	(−0.23, −0.03)*	−0.13	(−0.24, −0.02)*	−0.14	(−0.26, −0.01)*	−0.14	(−0.26, −0.02)*
County percent occupied housing units   with major/severe damage from   (per 5%)	0.02	(−0.09, 0.14)	0.03	(−0.08, 0.13)	0.04	(−0.05, 0.13)	0.03	(−0.06, 0.13)	0.04	(−0.06, 0.14)	0.04	(−0.06, 0.14)
Individual social cohesion scored	−0.78	(−1.25, −0.30)***					−0.17	(−0.51, 0.17)			0.01	(−0.37, 0.39)
Social support (vs. T3: High)
T2: Medium	1.36	(0.47, 2.25)**							0.81	(−0.04, 1.66)*	0.81	(−0.05, 1.67)*
T1: Low	1.77	(0.89, 2.66)***							1.34	(0.39, 2.30)**	1.35	(0.33, 2.37)**

^aSample limited to respondents who reported being of non-Hispanic black or white race and who did not change their county of residence after Hurricane Katrina. N=708; weighted N=1533.

^bAverage value over 25 imputations; −2 log likelihood for intercept-only model = 622.12; −2 log likelihood for model including only social cohesion, displacement, and their interaction = 554.86.

^cData from 2000 U.S. Census.

^dRange 0–4. Higher represents higher social cohesion.

^*p < 0.10;

^**p < 0.01;

^***p < 0.001

DISCUSSION

In a population-based sample of residents of the state of Mississippi 18–24 months after Hurricane Katrina, we found that joint exposure to post-Katrina displacement and low county-level

social cohesion was associated with a much greater likelihood of depression while exposure to just one of the risk factors was associated with only slightly higher odds of depression. The associations were evident after taking into account individual- and county-level covariates, including county-level Katrina exposure, and robust to additional adjustment for individually perceived social cohesion and social support.

Our results were consistent with existing evidence relating displacement to poorer mental health, as well as with past studies linking depression with the community social environment [6,21]. More generally, we addressed a previously identified gap in the epidemiology literature by applying the question of community social environmental influence on mental health to a postdisaster context [40]. Although the importance of social ties to postdisaster mental health has been widely acknowledged and empirically supported [3,14,41], research in this field has focused almost exclusively on aspects of the smaller-scale social environment, such as individual social ties and social support, rather than on the broader community-scale social environment. In one study that considered the broader social environment, Wind et al. [42] found in a flood-affected community in England that individually perceived social capital was inversely associated with depression 11 months after the flood. In another, Ahern and Galea [43] hypothesized that associations between higher neighborhood-level income inequality and higher odds of depression among low-income residents six months after a terrorist attack might be explained in part by differences in social interactions between residents of neighborhoods with different inequality levels.

The strong interaction we observed between displacement and community social cohesion may help explain inconsistencies across previous studies in associations between displacement and depression [6,7]. Furthermore, our results suggest a need for further research into how different social and individual conditions interact to shape longer-term mental health after a traumatic event. Joint exposure to risks at the individual, interpersonal, and community levels may result in vulnerabilities that are not evident when each of the risks is considered separately. For example, in a study of Katrina survivors displaced to Colorado, Wadsworth et al. [44] found that different patterns of coping were predictive of different trajectories of depression; these patterns may in turn be differentially influenced by the community social environment.

In addition to consideration to the joint influence of post-trauma experiences, further research that more explicitly accounts for time will be important for fully characterizing how displacement and social cohesion relate to depression after a disaster. In our analysis, estimates for the main-effect association of displacement and the joint association of displacement and low social cohesion were larger for past-month depression than for any depression since Hurricane Katrina while the main-effect estimate for low social cohesion was smaller for past-month depression. Although care should be taken in interpreting these differences given the substantial overlap in confidence intervals for the two outcomes, they suggest that ongoing stressors related to displacement may confer risk for depression that either endures over the longer term or has a relatively late onset relative to the time of the traumatic event, and that this longer-term risk can be compounded by the community environment. In addition, there is already evidence that disasters such as Hurricane Katrina themselves affect the social environment [17]. Therefore, it will be important to understand how social cohesion levels before a disaster are affected by the disaster itself and, in turn, how pre- and postdisaster social cohesion may act in concert to affect the mental health of community members. Research is also needed to examine the effects on mental health of the multiple social environments to which displaced populations are exposed as they relocate in the period after a disaster. Finally, research is needed that addresses permanently relocated populations, among whom the potential benefits of a cohesive community must be balanced against the stresses of adapting to a new community [6].

Because of the cross-sectional nature of this study we cannot rule out the possibility that the onset of depression preceded exposure to a given level of social cohesion or that depressed residents were more likely to leave their homes following the disaster, precluding a causal interpretation of our results. Similarly, we cannot rule out the possible selection of individuals predisposed to depression into counties with low social cohesion. This is a particularly important consideration with respect to the construction of our social cohesion measure, as it is possible that individuals’ mental health may influence their reports of social cohesion. That said, the robustness of our results to adjustment for individual psychological characteristics and perceptions (social cohesion, social support, and prior PTSD symptoms) suggest that further research into a potential causal link between social cohesion and postdisaster depression among populations grappling with disaster-related stressors is warranted. Inherent to this analysis is also the assumption that the average of individually reported social cohesion levels represents a valid estimate of the true level of social cohesion in each county. This is a substantial assumption, but it is supported by our population-based sample design and is in keeping with previous research of the area-level social environment and depression [21]. Finally, our estimates are imprecise, likely due in part to the relatively small number of counties and their limited range of social cohesion.

Our results highlight the importance of addressing the multiple, simultaneous disruptions that are a hallmark of mass traumatic events in order to identify vulnerable populations and understand the psychological ramifications of these events. They also demonstrate the utility of applying a socioecological model of mental health to post-event contexts. Just as mass traumatic events affect entire communities, community-level characteristics and processes may help minimize their initial damaging effects on individual members and promote recovery in the longer term. It is our hope that future research will identify attributes of communities that are beneficial to their members’ mental health after a traumatic event and elucidate the mechanisms through which these attributes operate.