Rasch Analysis and Differential Item Functioning of a Social Support Measure in Jail Inmates With HIV Infection

Abstract

The protective effects of social support on health have been documented in a variety of groups. For HIV-infected persons released from correctional settings, strong social support may be particularly important for obtaining effective postrelease medical treatment and supportive services. Researchers and program evaluators seeking to improve access and adherence to postrelease HIV medical care in this population need accurate measures for the level and type of social support, but current measures have not been fully validated for incarcerated individuals with HIV infection. We used the Rasch model to test the Medical Outcomes Study (MOS) social support survey. Data for the analysis were collected as part of the EnhanceLink project in the five urban jails where the MOS was administered. Findings indicate that the MOS survey items may not capture the entire variability of person abilities. Respondents showed problems in discriminating among response options, indicating potential systematic bias. In addition, while there was no significant gender difference, overall levels of social support differed by gender. Further research is warranted to develop more effective social support measurement tools that can better guide interventions for persons transitioning from jail and prison to the community.

Keywords

social support Rasch model inmates gender measurement

Introduction

The Significance of Social Support Among Inmates With HIV Infection

The rate of incarceration in the United States has increased exponentially since 1980 (Sabol & Couture, 2008), and more than 2.2 million individuals currently are held in jails and prisons on any given day (Glaze, 2011). Incarcerated individuals disproportionately experience myriad socioeconomic, behavioral, and health problems, including HIV infection (Dwyer, Fish, Gallucci, & Walker, 2011; Hammett, Harmon, & Rhodes, 2002). The prevalence rate of HIV among people in prisons is about 2.5 times higher than in the general population (AIDS.GOV, 2012; Wakeman & Rich, 2010). Similarly, individuals with HIV infection are more often involved in the criminal justice system compared with the general population: annually about 14% of persons with HIV are incarcerated, compared to less than 3% of the general population (Spaulding et al., 2009; Institute for Criminal Policy Research, 2016).

Studies have documented that the experiences of both HIV infection and incarceration have detrimental effects on psychosocial conditions, physical and mental health outcomes, and quality of life (Burgoyne, 2005; Crystal et al., 2003; Galvan, Davis, Banks, & Bing, 2008; Giordano et al., 2007; Larios, Davis, Gallo, Heinrich, & Talavera, 2008; Lichtenstein, Laska, & Clair, 2002; McCoy et al., 2009; Wohl et al., 2011). These effects may be countered by having strong social support systems, which have been shown to improve access and adherence to care (DiMatteo, 2004; Gallant, 2003; Heaney & Israel, 1996; Parkerson, Broadhead, & Tse, 1995), and result in better health outcomes for those with HIV infection (Crystal et al., 2003; Kessler & McLeod, 1985; Larios et al., 2008; Ryan, Huebner, Diaz, & Sanchez, 2009; Sherbourne, Meredith, Rogers, & Ware, 1992; Victor, Bowling, Bond, & Scambler, 2003). In the case of inmates, social support can also help them navigate difficult life circumstances, as they return to the community (Mears, Mancini, Gertz, & Bratton, 2008) and avoid reengagement in criminal behavior (Hyman, Gold, & Cott, 2003; Knudsen et al., 2008; Pratt & Godsey, 2006; Staton-Tindall, Royse, & Leukefeld, 2007).

Stigma and social isolation associated with HIV have been shown to affect patients’ treatment adherence (Rintamaki, Davis, Skripkauskas, Bennett, & Wolf, 2006; Simoni, Frick, & Huang, 2006). Social support can function as a coping resource that buffers the effects of stressful events and helps manage the effects of stigma and isolation (Dilley, Ochitill, Perl, & Volberding, 1985; Fleishman et al., 2000; Lichtenstein et al., 2002; Tross & Hirsch, 1988). Social support networks that are limited due to stigma and isolation may be one mediating effect.

Although social support has been defined and operationalized in a variety of ways (Thoits, 1982), social support generally indicates “support accessible to an individual through social ties to other individuals, groups, and the larger community” (Lin, Simeone, Ensel, & Kuo, 1979, p. 109). Social support exists at multiple levels of social relationships including individual-level interactions with family and friends, neighborhood-level social ties, and institutional-level formal relationships, such as interactions with social services and health-care providers (Orrick et al., 2011). Social support networks play a critical role in helping inmates make successful transitions from corrections to the community (Fontaine, Gilchrist-Scott, Denver, & Rossman, 2012; La Vigne, Shollenberger, & Debus, 2009; Rose & Clear, 2001), but many inmates may have exhausted support from family and friends due to long histories of drug use and involvement with criminal activities (Lichtenstein et al., 2002; Nichols et al., 2002). Consequently, the types of social relations common among persons living with HIV who have been incarcerated and are returning to the community may not be fully described by the domains typically used to measure social support.

Because of multiple intersecting burdens, individuals with HIV infection who are released from correctional settings have unique challenges in reestablishing support relations (Fontana & Beckerman, 2007; Nunn et al., 2010). Furthermore, while social support is generally positively associated with better health outcomes, support from the some of the social networks that incarcerated individuals may return to upon reentering the community can also lead to high-risk behavior (Falkin & Strauss, 2003; Latkin et al., 2009; Uchino, 2006).

One limitation of conventional social support studies is that they are less effective in capturing unique characteristics of support networks among subgroups of people who are predominantly excluded from mainstream social relations (Coyne, Wortman, & Lehman, 1988; Shinn, Lehmann, & Wong, 1984). Of particular relevance for this study, incarceration weakens social relations of not only those with a history of incarceration but also their families and communities (Rose & Clear, 2002; Morenoff & Harding, 2014; Thomas & Sampson, 2005). Thus, existing social support measures may not be well suited to characterize social support among imprisoned individuals with HIV infection. For example, Kim and Mazza (2013) used a Rasch model to evaluate social support scores among incarcerated women in a county jail and found that social support measurement items were not evenly distributed across person abilities, indicating inadequate variability in the item difficulty (Kim & Mazza, 2013). The current study expands Kim and Mazza’s analysis to evaluate the use of social support measures for persons affected by both incarceration and HIV infection.

Rasch Model and the Medical Outcomes Study Social Support Survey (MOS-SSS)

Although the MOS-SSS has been used and validated in a variety of groups, this measure has not been thoroughly evaluated for use with individuals with HIV infection in correctional settings. Likewise, while research has shown that female and male inmates experience different life challenges and barriers to accessing social and health services (Binswanger et al., 2010; Williams et al., 2013), gender differences in the use of the instrument with this population are not well documented.

We applied the Rasch model to examine the characteristics of social support and the differences in the use of the measurement among men and women with HIV infection who are incarcerated in jails (Hays, Sherbourne, & Mazel, 1994; Sherbourne et al., 1992; Sherbourne & Stewart, 1991). The MOS-SSS measures four dimensions of social support: emotional and information, tangible, affectionate, and positive social interaction. A higher score indicates a higher level of social support (Hays et al., 1994; Sherbourne, 1988).

One of the strengths of the Rasch model is that it seeks to specify data that fit the measurement, rather than to fit the model to account for the data. Thus, fit statistics in Rasch measurement summarize the discrepancies between the Rasch model specifications and the empirical data (Bond & Fox, 2007). Furthermore, the Rasch model estimates item locations on the scale of how rare or frequent an item is to be endorsed by respondents, known as “item difficulties”(Andrich, 1988; Bond & Fox, 2007), which can help determine how items are used differently by different groups and across different contexts. Such group differences in the use of a measurement (i.e., differential item functioning [DIF]) may introduce biases when comparing the overall scores of the instrument (Facon & Nuchadee, 2010; Lee, Peterson, & Dixon, 2010).

For individual respondents, probabilistic estimates of person-level severity are obtained based on their endorsement of items. Person severity level assumes that respondents who are able to endorse more difficult items should also be able to endorse easier items as well, thus would score high on the overall measurement scale. For example, students who can solve multiplication and division problems should also be able to solve addition or subtraction problems. Items and persons that do not fit this pattern are considered to be misfitting and thus require further evaluation (Bond & Fox, 2007). Items that are ambiguously worded or do not reflect underlying constructs of a particular measure may cause misfit, which often can be addressed by improving the clarity of questions or by excluding them from the measure.

Specific Aims

Because of the ability to compare different response patterns between groups, the Rasch model has been applied in a broad range of health-related studies. The reliability and validity of the MOS-SSS have not been established for individuals with HIV infection in correctional settings. Gender differences in the item response and severity (frequency) of the measurement also need further evaluation. Thus, we take advantage of the strengths of Rasch analysis to (1) explore responses to the MOS-SSS when administered to a persons detained in jail who are living with HIV and (2) compare item responses between men and women in this sample.

Method

Data Source

We used data collected by the EnhanceLink project, a Special Project of National Significance funded by the HIV/AIDS Bureau of the Health Resources and Services Administration and comprised of 10 jail-based demonstration programs (Draine et al., 2011). Individuals with confirmed HIV were eligible for the evaluation and usually enrolled while still in jail. Most EnhanceLink participants (91%) had been diagnosed with HIV for more than 2 years and 81% reported having received antiretroviral therapy. However, only 49% of EnhanceLink participants had engaged in antiretroviral therapy in 7 days before being arrested (Stein et al., 2012).

The baseline survey included an optional module developed by Sherburne and Steward (Hays et al., 1994; Sherbourne & Stewart, 1991) and adapted by Huba (Huba & Melchior, 1996) to examine social support. Five sites administered the social support module: Chicago, IL; Cleveland, OH; Columbia, SC; New Haven, CT; and Philadelphia, PA. Four of the five sites collected client-level data on men and women, while the Chicago program exclusively targeted women.

While sites varied in criteria for enrollment in the client-level evaluation, all sites limited participation to adult inmates 18 years of age or older. A total of 1,270 individuals were eligible and consented to the 10-site EnhanceLink evaluation. A baseline interview (conducted in English or Spanish) was conducted generally by project staff who completed a web-based training on survey implementation. A detailed description of the EnhanceLink study is provided elsewhere (Draine et al., 2011). The five sites that administered the social support module accounted for 411 of these participants, and data for the current analysis were drawn from these baseline assessments.

Variables

The MOS-SSS quantified the respondent’s level of social support before the current incarceration. We used the original MOS-SSS that consists of four subscales and one additional functional social support item. The Emotional and Informational Support subscale is composed of 8 items, the Tangible Support subscale has 4 items, and the Affectionate Support and Positive Social Interaction subscales each consists of 3 items. The 19 items were rated on a 5-point response scale, ranging from 0 to 76. Response categories included Category 0 (none of the time), Category 1 (little of the time), Category 2 (some of the time), Category 3 (most of the time), and Category 4 (all of the time). We examined the four subscales as well as the overall MOS-SSS score, which was calculated using the formula suggested by the RAND Corporation.

Demographic variables in this analysis included age, gender (female and male), race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and other), education, employment status (unemployed, part time, and full time), and marital status (never married, married, divorced/separated, and widowed). Transgender participants (two cases) were excluded from the analysis due to their low number. Age was treated as a continuous variable. A dichotomous variable for high school education or GED was created from the original education variable. For employment status, we recoded “days you were paid in the 30 days before your most recent arrest” into unemployed (0 days), part-time employed (1–20 days), and full-time employed (more than 21 days).

Comorbidities and risk factor variables examined in the analysis were total number of incarcerations, length of stay for the current incarceration, mental health conditions, and substance use. Total number of incarcerations and the length of stay were continuous variables. Mental health conditions were measured with a series of questions asking history of mental health treatment for psychological or emotional problems, being on a psychiatric disability pension, history of depression, anxiety, tension, hallucinations, trouble-controlling violent behavior, thoughts of suicide, and being on medications for psychological or emotional problems. We then created two dummy variables indicating lifetime mental health problems when any psychological or emotional problems were endorsed (coded 1 = any and 0 = none). Similarly, we created a dummy variable indicating lifetime history of substance use, when any of the following were endorsed use of heroin, methadone, buprenorphine or suboxone, other opioids, barbiturates, sedatives or tranquilizers, cocaine, methamphetamine/amphetamines, cannabis, hallucinogens, and inhalants.

Analysis

Descriptive statistics

First, we described characteristics of the sample and compared MOS-SSS, demographics, and comorbidities/risk factors by gender. Descriptive statistics, t-tests, and χ² tests were used to compare men and women.

One of the properties of Rasch analysis is unidimensionality of a measurement. We performed exploratory factor analysis to determine whether to proceed with the Rasch model. As shown in Table 1, principal component analysis (PCA) indicated that the MOS-SSS met the unidimensionality requirement, thus we continued with the Rasch analysis.

Table 1.

MOS Factor Components, Factor Loading, and Subscale Domains.

Subscale	Factor Loading	Rasch Item Measure
Subscale	Factor Loading	Male	Female	All
Emotional/Information Support
1. Someone you can count on to listen to you	.581	.15	.08	.12
2. Someone to give you information	.769	−.32	−.14	−.25
3. Someone to give you advice about a crisis	.673	−.15	−.02	−.10
4. Someone to confide in or talk about your problems	.654	.33	.28	.31
5. Someone whose advice you really want	.669	−.47	−.54	−.49
6. Someone to share your most private worries/fears	.721	.00	.03	.00
7. Someone to turn to for suggestions for problems	.758	−.12	.10	−.04
8. Someone who understands your problems	.761	−.01	−.01	−.01
Tangible support
9. Someone to help you if you were confined to bed	.703	−.03	−.16	−.08
10. Someone to take you to the doctor if you needed it	.655	.11	.23	.16
11. Someone to prepare your meals if you were unable	.632	−.02	.01	−.02
12. Someone to help with daily chores if you were sick	.713	.00	.00	.00
Affectionate Support
13. Someone who shows you love and affection	.712	.11	.02	.08
14. Someone to love and make you feel wanted	.653	.34	.13	.26
15. Someone who hugs you	.732	.17	.26	.20
Positive social interaction
16. Someone to have a good time with	.768	.08	.18	.12
17. Someone to get together with for relaxation	.730	.09	−.03	.05
18. Someone to do something enjoyable with	.762	.12	.05	.09
Additional item
19. Someone to do things with to get mind off things	.714	−.36	−.44	−.39

Rasch analysis

We then conducted Rasch analysis with Winsteps (Version 3.72.0) statistical software (Linacre, 2007) to obtain linear interval measures. The Rasch model estimates the log odds of a respondent choosing a particular response category (trait-level estimate) for an item (Bond & Fox, 2007, pp. 281–282). Rating scale categories are ordered response options, where choosing a response category of an item can be considered choosing the particular response category over the one below (Litz, Penk, Gerardi, & Keane, 1991). For example, in the MOS-SSS measurement, one may pass the threshold of choosing “some of the time,” which would be rated 3, and instead choose response category “most of the time,” which would be rated 4. The Rasch Rating Scale Model (RSM) was used to provide additional insight about the measure, regarding how individual items function, and how respondents utilize the 5-point response scale.

Response category analysis

The monotonicity assumption of the Rasch model indicates that with the increase in person ability, the probability of being correct on an item also monotonically increases (Bond & Fox, 2007). We examined the item response function and compared the item-person response results between male and female respondents.

PCA

PCA of Rasch item residuals examines residual variance that was not accounted for by the Rasch analysis (Bond & Fox, 2007). PCA results are then used to estimate the unidimensionality of items. When variance explained by the measurement dimension is greater than 40%, the unidimensionality requirement is considered to have been met (Reckase, 1979). Unexplained variance in the first contrast of the data should be less than 15% for a rival factor. In addition, local independence of items was examined. Residual bivariate correlations among the items were reviewed. Items with residual correlations greater than 0.4 are considered to be highly correlated, which signifies that the local independence assumption was not met.

Item quality

Conventionally, mean square fit statistics between 0.75 and 1.33 logits are considered appropriate (Wilson, 2005). In this study, items were regarded as misfitting, if the mean squares of in-fit or out-fit were higher than 1.33 or lower than 0.75. These criteria are appropriate even for large samples (Bond & Fox, 2007). Fit statistics for both items and persons indicate the level of deviation of observed responses from expected response distributions. The Rasch model uses the person and item totals to estimate person ability and item difficulty, and these values are used to estimate the expected responses. An observed response should be associated with a high probability of that response, and the deviation of the observed response from the probabilistic expectation is an indicator of misfit. Such indices may reflect a poor item. In addition, items that are endorsed by most respondents might not be useful in providing information about the construct or the respondents. On the other hand, to differentiate person ability or item difficulty, a measurement needs to include easy and difficult items that can be used to detect differences among both very low-scoring and very high-scoring responses. The lack of such variability among items may result in floor or ceiling effects.

Another useful diagnostic provided by the person/item maps of the Rasch model is that of how well the items are centered on the population of interest. Overall, the Rasch model is designed so that a 50% probability of success for any person on an item is set as zero on the logit scale (log of odds of succeeding). Thus, negative logit scores indicate items being more difficult than average (less than 50% chance of succeeding), and positive logit scores indicate items being easier than average (more than 50% chance of succeeding). On the logit scale, the difference between 0 and ±1 logit is roughly a difference of ±23% from the 50% probability; the difference between −2 and +2 logits is roughly the difference between 12% chance of succeeding and 88% of chance of succeeding (Winsteps). The items are considered to “fit” the population of interest when the difficulty of items increases by a roughly one-unit increment on the logit scale. This outcome would be an indication that the items are appropriate for the target population. Additionally, items should have an appropriate spread that ranges across the span of persons measured to capture the wide range of variability of person abilities on the construct (Bond & Fox, 2007). Using this approach, the item maps created by the Rasch model for male and female inmates were then compared.

Reliability

Person/item reliability was used. Rasch person reliability, which is equivalent to Cronbach’s α, is expected to meet the 0.80 criteria. Item reliability, which is a measure of reproducibility of “item placements” (Bond & Fox, 2007, p. 41) when same items were used for other same size samples, should also be above the 0.80 criteria. We also examined the item and person separation. The item separation index details the number of standard errors (SEs) of spread across the items (Bond & Fox, 2007). Using the item separation index, along with the item reliability score, we estimated the ability of the measures to position the items on the frequency hierarchy.

Validity

We examined construct validity for the MOS-SSS using Rasch item hierarchy provided by the item difficulty estimates. We illustrated the items of a scale hierarchy, placing frequently endorsed items on the vertical difficulty scale at the bottom and less frequently endorsed items on top.

DIF

Finally, gender differences in the responses to MOS were compared. As Bond and Fox note, the Rasch model requires that relative item estimates (i.e., item difficulty estimates) remain invariant across subgroups of persons (e.g., females and males; Bond & Fox, 2007). DIF examines whether items have significantly different meanings for different groups. Items that show significant difference between groups should be investigated to examine whether the groups in comparison interpret the underlying construct differently and whether other characteristics of groups may contribute to the group difference.

A significant DIF contrast was based on ≥0.55 logit difference for the gender comparison which is approximately half a standard deviation (SD) for the person measure (Conrad, Dennis, Bezruczko, Funk, & Riley, 2007; Norman, Sloan, & Wyrwich, 2003). Conventionally, half of a SE is a common criterion for clinical significance (Conrad et al., 2007; Norman et al., 2003 ). Because SE is normalized, the minimum possible SE is 0.20 logit. And, with SE of 0.20, a significant DIF contrast needs to be ±0.55 to be statistically significant (Gantschnig, Page, & Fisher, 2011).

Results

Descriptive Characteristics of the Sample

Table 2 describes demographic characteristics of this sample of jail detainees living with HIV (N = 411). The majority of the sample was male, African American, not married and typically unemployed in 30 days before the current incarceration. Median age was 41 years with a range of 19–63 years. Most participants reported past drug use, a history of mental health problems, and multiple incarcerations. Less than one half of the participants had at least a high school education or GED, though 53% of the sample was missing information on education level.

Table 2.

Demographics Characteristics of the Sample.

Variable	Total (N = 411)	Male (n = 256)	Female (n = 155)	p
Race/ethnicity
White	15.9	13.0	20.6
African American	66.5	66.1	67.1	0.04
Hispanic	4.9	6.3	2.6
Other	12.7	14.6	9.7
Age	40.9	41.1	40.6	0.55
Marital status
Single	48.3	52.5	32.3
Married	8.5	8.2	9.0
Separated/divorced	13.7	14.1	12.9	0.12
Widowed	2.9	2.7	3.2
In relationship	26.6	22.4	33.5
Living with a partner	24.9	21.6	30.3	0.05
Education
High school/GED or more	47.2	57.9	31.6	0.001
Employment
Unemployed	79.2	79.6	79.2	0.49
Part time	12.7	12.7	14.6
Full time/student	7.6	7.6	6.1
Ever used drug	83.3	77.8	91.2	0.001
Ever mental illness	87.4	85.4	90.9	0.07
Social support (total)^a	49.1	46.8	52.9	0.01
Tangible	9.9	9.4	10.8	0.01
Emotional	20.7	19.8	22.1	0.05
Affectionate	8.3	7.9	9.0	0.01
Positive interaction	7.6	7.3	8.2	0.02
Incarceration history
Total number of arrests	18.0	15.0	22.0	0.03
Total years incarcerated
<2 years	39.2	36.0	43.5
2-5 Years	24.7	19.8	31.5	0.002
>5 years	36.1	44.2	25.0

Note. N = 411.

^aAvailable Social Support Scale ranges from 0 to 76.

Overall, male inmates were significantly more likely to have a high school/GED or greater education, to report being single, and to have incarcerations totaling more than 5 years. Women were significantly more likely than men to report living with a partner, to be unemployed, to have ever used drugs, and to have a greater number of arrests. Among lifetime substance users, women (91%) were also more likely than men (78%) to have been in a drug treatment program (p < .01).

The average MOS-SSS score of the study sample was 49.1 (SD = 24.3), ranging from 0 to 76 and was higher for female inmates (52.9) compared with male inmates (46.8; p < .01). Women also had higher social support scores for all four subscales (Table 1), and the differences were most pronounced for Emotional or Informational and Affectionate Supports. Furthermore, the mean Rasch measure was higher for women (0.89) compared with the mean for men (0.40). This finding indicates that positively endorsing items on the MOS-SSS was easier for women compared with men.

Table 3 shows the results of linear regressions that we modeled to explain the level of social support. Although it was not a primary aim of this study, we felt that identifying factors associated with levels of social support in this population could help us better interpret item response patterns of the measurement tool. Overall, being female and living with a partner were consistently associated with a higher social support level. Interestingly, however, being employed, having high school or greater education, time spent incarcerated, and having a history of drug use or mental health problems were not associated with social support, in both univariate and multiple regressions.

Table 3.

Linear Regression Models Explaining Social Support.

	Standardized Regression Coefficients
Predictors	Model I	Model II	Model III	Model IV	Model V
Female	0.11*	0.18*	0.32**	0.30**	0.28**
African American	0.10*	0.04	0.14	0.11	0.09
Age	−0.08+	0.002	−0.04	−0.12	−0.06
Living with a partner	—	0.28**	0.27**	0.27**	0.27**
Employed	—	0.13⁺	0.06	0.12	0.09
≥High school education	—	0.06	0.15	0.11	0.10
Number of incarcerations	—	—	0.08	—	—
Years in jail/prison	—	—	—	—	—
<2 years	—	—	—	−0.08	—
>5 years	—	—	—	0.09	—
Ever drug use	—	—	—	—	−0.07
Ever mental illness	—	—	—	—	−0.04

Note. N = 411.

*p < .05. **p < .01. ⁺ p < .10.

Dimensionality

Table 1 summarized the factor solution for the MOS. While the original MOS measurement consists of four distinct subdomains and one additional item, the factor analysis results showed only one component, with the initial eigenvalue of 13.4 explaining 70.31% of variance. Factor loadings of all items, which were treated as continuous variables, ranged from 0.58 to 0.77.

Results of the PCA showed that the variance explained by the measurement dimension was 57.1%, much greater than the 40% recommendation by Reckase (1979). The unexplained variance (5.6%) in the first contrast of the data fell below the criterion of 15% for a rival factor (i.e., a competing factor). This finding further confirmed that the measure is unidimensional and appropriate for analysis via the Rasch model. Residual correlations among 19 items ranged from −0.31 to 0.33, which indicated that the local independence assumption has been met.

Rasch Analysis: MOS-SSS

We primarily used the RSM, given that the MOS-SSS is designed with a 5-point Likert-type scale. In addition, we utilized the partial credit model (PCM) to examine potential disordered items in the MOS-SSS. The PCM results showed that none of the items had inconsistent or reversal category orders. Similarly, we ruled out the graded response model, which is an alternative to PCM, to examine ordered item categories of polytomous data (Muraki, 1990; Thissen & Steinberg, 1986). More importantly, the MOS instrument has been used as a Likert-type scale measurement tool that assumes a uniform number of thresholds for all items, thus implying that all items share the same rating scale structure. Given these factors, we proceeded with the RSM approach for the remainder of the analysis.

The Rasch item map depicts person ability and item difficulty on the same logit scale. The item mean is fixed at zero. The person mean on the measure is 0.59 logits, indicating that for this population, the social support score was higher than the average mean of the items. We found that many respondents scored very high or very low on the social support measure. In addition, the difference between the mean of the items and the mean of the persons was within one logit, suggesting that the items fit the population of interest. However, the items did not have an even spread across the span of persons measured. This finding indicates that the MOS-SSS item difficulties may not vary enough to adequately capture the full range of person ability. Person difficulty ranged between −4.87 and 4.97. But item difficulty was much more restricted, ranging from −0.49 to 0.31. This highly limited range of item difficulty indicates that the measurement items do not cover a full range of MOS levels in the study population, which results in the large proportions of respondents scoring the highest and the lowest limits (ceiling and flooring effects).

In particular, 78 respondents endorsed the highest response for all items. On the other hand, 32 respondents endorsed the lowest response category for all items. Since these extreme cases may skew Rasch reliability, we present Rasch summary statistics for all cases and for the nonextreme (nonceiling, nonfloor) cases in Table 4.

Table 4.

MOS Rasch Model Summary Statistics.

Sample Size	Male	Female	All
Sample Size	266	163	429
Nonextreme cases only
Mean person measure	0.40	0.89	0.60
SD	1.29	1.31	1.31
Range	−3.71 ∼ 3.76	−3.66 ∼ 3.86	−3.94 ∼ 3.79
Person reliability	0.94	0.92	0.93
Person separation	3.82	3.38	3.38
Item reliability	0.90	0.90	0.90
Item separation	3.00	3.00	3.00
All cases
Mean person measure	0.86	1.18	0.99
SD	2.72	2.40	2.60
Range	−4.94 ∼ 4.92	−4.83 ∼ 5.05	−4.97 ∼ 4.87
Model reliability	0.86	0.86	0.94
Correlation, raw score	0.94	0.93	0.98
χ²	8,366.43	5,334.33	13,646.20
p	.01	.01	.01
df	3,392	2,304	5,683
Unidimensionality
Variance explained	24.9 (56.7%)	23.6 (55.4%)	27.6 (57.4%)

Note. MOS = Medical Outcomes Study.

Reliability

Table 4 describes Rasch reliability of the MOS-SSS. The person summary statistics yielded a person reliability of 0.93. The item summary statistics revealed an item reliability of 0.90, which was comparable to the reliability from the original RAND study in the general population (α = .91; Sherbourne & Stewart, 1991).

Validity

In terms of construct validity, the items of the MOS form a hierarchy with most commonly endorsed items on the bottom and less endorsed items on the top. The more frequently endorsed items were “someone to love and make you feel wanted” (Item #14) and “someone who shows you love and affection” (Item # 13), which were from the subscale Affectionate Support. This finding demonstrates that affectionate support was more often endorsed among incarcerated individuals. The least endorsed items were “help with daily chores if you were sick” (Item #12) and “take you to the doctor if you needed it.”

We also examined category structures overall and for each item. One item, “someone you can count on to listen to you,” was shown to be misfitting (Infit MNSQ = 1.38) according to the Wilson criterion (which was between 0.75 and 1.33; Linacre, 2007; Wilson, 2005). Step calibration results also indicate that this item is disordered, that is, a higher category number could correspond to a lower level of support. As discussed earlier, the Rasch requirement is that average measures (endorsements) advance monotonically and that step calibrations between responses (e.g., from 0 = none of the time to 1 = little of the time and to 3 = most of the time) increase monotonically as well (Bond & Fox, 2007). A disordered category indicates that respondents are not using the rating scale as the researcher intended (Linacre, 2007).

Figure 1 illustrates category probabilities, which are the probability curves for the five response categories regarding item difficulty. Overall, Category 1 (little of the time) and Category 3 (most of the time) seem to be depressed, which indicates that these categories were less utilized by the respondents, compared with Category 0 (none of the time), Category 2 (some of the time), and Category 4 (all of the time).

Figure 1.

Response category probability curve.

DIF

We explored item difficulty by gender. The items are sorted using all cases, from the easiest item to the hardest items. We then compared female and male item difficulties. In this study, the significant DIF contrast was based on ≥0.55 logit difference (Conrad et al., 2007; Gantschnig et al., 2011; Norman et al., 2003 ), and we found that none of the items was significantly different by gender. This finding indicates that female and male study participants did not statistically differ in the use/interpretation of social support items. Consequently, the mean difference of MOS-SSS by gender (Table 2) reflects the actual difference in the perceived level of social support rather than a difference in the item difficulty by gender.

Discussion

Overall, the MOS-SSS measure was highly reliable: conventional Cronbach’s α and the Rasch reliability statistic were both greater than 0.9. This finding reflects previous studies of MOS-SSS in a variety of population groups (Hays et al., 1994; McDowell, 2006; Reckase, 1979; Sherbourne & Stewart, 1991).

Our findings suggest several issues, however, concerning the measurement’s validity when used with jail detainees living with HIV. First, potential systematic bias in participants’ responses to the measurement was observed. In general, if more than 15% of responses result in the minimum or maximum scores, ceiling or floor effects need to be further evaluated (McHorney & Tarlov, 1995). In our study, over 25% of respondents who answered the MOS-SSS reported full or no social support in the community prior to their incarceration. Such ceiling and floor effects could have happened because the measurement did not appropriately capture or represent the range of perceived social support. This outcome may mean that the MOS-SSS items do not sufficiently encompass varying degrees of social support in the study sample. The current items were shown to be clustered around the center and do not allow more detailed differentiation at the both extremes (thus, floor and ceiling effects). Additional items that are more difficult to endorse may help improve the performance of the measurement. Previous studies suggest that implementing a disease- or situation-specific MOS-SSS can reduce ceiling or floor effects (Bombardier et al., 1995; Shahriar, Delate, Hays, & Coons, 2003). In our study, the ceiling or floor effects might be minimized by presenting a specific condition(s) for which participants rate their social support. To the best of our knowledge, the MOS-SSS measurement has not been validated with incarcerated populations. It might be beneficial to examine whether social support is content-specific. For example, inmates may receive support to deal with health problems but may not have access to social support regarding issues around incarceration or substance use problems.

One reason for the ceiling and floor effects in our study sample might be related to different response styles and the factors influencing these styles (Bachman & O’Malley, 1984; Greenleaf, 1992; Van Vaerenbergh & Thomas, 2013). Racial and cultural differences in survey response patterns have long been documented (Bachman & O’Malley, 1984; de Jong, Steenkamp, Fox, & Baumgartner, 2008; Greenleaf, 1992; Van Vaerenbergh & Thomas, 2013), and African Americans and Hispanics are more likely to show a higher rate of extreme responses compared with Whites. However, reasons for racial differences are not well established.

In reviewing previous study results, Van Vaerenbergh and Thomas summarize that respondents with lower levels of education and income are more likely to show a higher rate of extreme responses (Van Vaerenbergh & Thomas, 2013). Morren and colleagues’ cognitive interview findings indicate that extreme responders tend to consider the item wording and weigh arguments less precisely (Morren, Gelissen, & Vermunt, 2013). Van Vaerenbergh and Thomas (2013) conclude that the current literature suggests that respondents who are less tolerant to ambiguity and prefer decisiveness are more likely to endorse extreme responses (Van Vaerenbergh & Thomas, 2013). Although it is outside the scope of our current study, it is possible that other personality traits of incarcerated individuals living with HIV may also influence item response patterns.

Our study participants are relatively less likely to have people who provide informational or tangible supports. While many of these participants likely require help to achieve adherence to appropriate HIV medical treatment as well as to address myriad social service needs after they are released from jail, this finding indicates they receive little such support. In fact, the level of social support for our sample was markedly lower (52.9) than in other groups reported in the research literature. For example, Sherbourne and Stewart reported a mean MOS of 70.1 for their 2,987 patients with chronic conditions sampled in three large cities in the United States (Sherbourne & Stewart, 1991). Kim and Mazza (2013) documented a mean MOS-SSS of 75 for incarcerated women in a local jail. Berard, VanDenKerkhof, Harrison, and Tranmer (2012) found that the mean MOS-SSS among older Canadian adults with heart failure was over 71. Closer to our findings, Bekele and others reported that a sample of mostly lower income Canadian adults with HIV infection scored 56.6 on MOS-SSS (Bekele et al., 2013).

The concept of formal and informal social support is an important issue concerning health-care interventions. Social support often is considered to be resources exchanged between individuals (Granovetter, 1973; Wellman & Wortley, 1990). However, those who are affected by multiple stigmatizing life events including HIV infection, incarceration, and substance use may have limited informal social support networks (Lichtenstein et al., 2002; Nichols et al., 2002), in part because they exhaust what social support is available (Blau, 1964; Domínguez & Watkins, 2003; Offer, 2012). Further, incarcerated persons living with HIV typically are from disadvantaged neighborhoods, and individuals in their social networks are likely to experience similar life conditions. The “homophily principle” in social networks may explain in part the limited resources and social supports available to incarcerated populations (Lin, 2001; McPherson, Smith-Lovin, & Cook, 2001).

Participants in our study appear to have very limited social networks from which to draw support. Given that a lack of social support diminishes access and adherence to appropriate HIV treatment, our findings suggest a need for formal services. That is, incarcerated individuals living with HIV often may lack sufficient informal social supports, thus organized formal support services may be particularly important for better health outcomes in this population. For the purposes of measurement in this case, the definition of social support should be expanded to include formal services. As alternative, related concepts such as social capital might be used to better understand forms of support in this population (Sampson, 2004; Sampson, Morenoff, Smedley, & Syme, 2000; Small, 2009).

Our findings suggest that the item “someone you can count on to listen to you” seems to be misfitting, disordered, and have a low discrimination estimate (0.45). Although not within the scope of the current article, this finding calls for further evaluation of the MOS-SSS and subsequent revision of the item (Andrich, 2002). In addition, two response categories were less utilized compared with the others. Specifically, Category 1 (little of the time) and Category 3 (most of the time) were observed less often. This finding indicates that respondents may have difficulty differentiating “little of the time” (Category 1) from “none of the time” (Category 0) and “most of the time” (Category 3) from “all of the time (Category 4). In the absence of clear definitions of these categories, it is unclear how respondents make decisions to choose one over the other. How respondents understand the concept being measured, compared with what was intended, needs to be further evaluated. Interpersonal incomparability, resulting in different ways to rate the level of social support, might also have affected the results (King, Murray, Salomon, & Tandon, 2004). Further studies, such as cognitive interviews, may help evaluate how this population understands and formulates their responses to social support questions. What respondents perceive as social support, and how they rate the level of social support needs to be explored.

While DIF showed no gender difference, there was a significant difference between men and women regarding the average Rasch measures. This finding was consistent with the descriptive summary that showed the average social support was higher for women than for men. However, additional regression analyses predicting social support showed that being female and living with someone were associated with higher levels of social support. Interestingly, even though substance use, incarceration, and mental health problems are known to negatively affect social relations and thus social support, women still seemed to receive more social support than men despite more often reporting these conditions. One possible reason for this outcome is that men and women draw social support from different social networks. Women might rely more on family members and relatives that are more forgiving, while men may tap into broader social relations. Further research is required to explore types of social networks and support by gender.

The temporal frame of responses to the measurement needs to be examined. First, it is not clear whether the respondents considered services and support that they received in jail or prison when responding to the questionnaire. Second, the time frame participants examined to answer the questionnaire is also problematic. Do respondents look back before their incarceration to rate their social support, or do they weigh what’s available at the moment? Detailed qualitative information regarding these issues can help refine not only the measurement tool but also the concept itself. In addition, the level of expected social support upon release may be important, particularly concerning successful reentry into the community. The comparison between inmates’ retrospective assessment of their social support and expected support levels seems to be a relevant research question for both measurement and service planning purposes.

A limitation of this study is a potential bias rooted in the high turnover rates of jails. Because a substantial proportion of jail detainees are released within a few days of their arrival, not all potential participants in the EnhanceLink study were enrolled. Those who were incarcerated long enough to participate in the study might be qualitatively different than those who were released quickly.

MOS-SSS is a self-reported measure of social support. Although it is beyond the scope of this study, whether perceived social support accurately reflects actual received support is an interesting research question. Furthermore, even if perceived support does not highly correlate with actual support, the effect of perceived support still seems to have positive effects on improved health and well-being (Nurullah, 2012). Nonetheless, the level of actual available support, particularly instrumental support, needs to be further evaluated. Studies have documented that individuals in disadvantaged social and economic conditions are more likely to exchange individual-level support, but when their network members are likely to be equally disadvantaged, support and resources available through these network members tend to be limited (Domínguez & Watkins, 2003; Quillian & Redd, 2006; Smith, 2000).

Conclusion

The findings warrant further investigation regarding how persons living with HIV who are in jail comprehend the concept of social support, retrieve information regarding social support, and construct their responses (Schuman, 2008). Different methods may produce different results that could be combined to better understand how this particular population comprehends social support. A better understanding of social support in this population would facilitate the tailoring of services to improve jail detainees’ ability to access appropriate HIV medical care and maintain that care after release from jail.

The overall MOS-SSS score tells only a partial story of social support in this vulnerable population. Reliability and validity of a measurement are also important areas of research for improving the health of vulnerable populations, yet the validity of key measurements often has not been assessed in subgroups that are substantially disadvantaged. Careful evaluation of instruments in such populations is necessary for developing more effective interventions, reducing disparities in health, and improving physical and behavioral health outcomes.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The Health Resources and Services Administration (H97HA08534).

References

AIDS.GOV. (2012). HIV/AIDS and incarceration. Retrieved from http://aids.gov/federal-resources/policies/incarceration/

Andrich

(1988). Rasch Models for measurement. Newbury Park, CA: Sage.

Andrich

(2002). Understanding Rasch measurement: Understanding resistance to the data-model relationship in Rasch’s paradigm: A reflection for the next generation. Journal of Applied Measurement, 3(3), 325–359.

Bachman

O’Malley

(1984). Black-White differences in self-esteem: Are they affected by response styles? American Journal of Sociology, 90, 624–639.

Bekele

Rourke

Tucker

Greene

Sobota

Koornstra

, … Positive Spaces Healthy Places Team. (2013). Direct and indirect effects of perceived social support on health-related quality of life in persons living with HIV/AIDS. AIDS Care, 25, 337–346.

Berard

VanDenKerkhof

Harrison

Tranmer

(2012). Gender differences in the influence of social support on one-year changes in functional status in older patients with heart failure. Cardiology Research and Practice, 2012, 1–10.

Binswanger

Merrill

Krueger

White

Booth

Elmore

(2010). Gender differences in chronic medical, psychiatric, and substance-dependence disorders among jail inmates. American Journal of Public Health, 100, 476–482.

Blau

(1964). Exchange and power in social life. New York, NY: Wiley.

Bombardier

Melfi

Paul

Green

Hawker

Wright

Coyte

(1995). Comparison of a generic and a disease-specific measure of pain and physical function after knee replacement surgery. Medical Care, 33, AS131–AS144.

10.

Bond

T. G.

Fox

C. M.

(2007). Applying the Rasch model: Fundamental measurement in the human sciences (2nd ed.). Mahwah, NJ: Lawrence Erlbaum.

11.

Burgoyne

(2005). Exploring direction of causation between social support and clinical outcome for HIV-positive adults in the context of highly active antiretroviral therapy. AIDS Care, 17, 111–124.

12.

Conrad

Dennis

Bezruczko

Funk

Riley

(2007). Substance use disorder symptoms: Evidence of differential item functioning by age. Journal of Applied Measurement, 8, 373–387.

13.

Coyne

Wortman

Lehman

(1988). The other side of support: Emotional over involvement and miscarried helping. In Gottlieb

(Ed.), Marshaling social support: Formats, processes, and effects (pp. 305–330). Thousand Oaks, CA: Sage.

14.

Crystal

Akincigil

Sambamoorthi

Wenger

Fleishman

Zingmond

(2003). The diverse older HIV-positive population: A national profile of economic circumstances, social support, and quality of life. Journal of Acquired Immune Deficiency Syndromes, 33, S76–S83.

15.

de Jong

Steenkamp

J.-B. E.

Fox

J.-P.

Baumgartner

(2008). Using item response theory to measure extreme response style in marketing research: A global investigation. Journal of Marketing Research, 45, 104–115.

16.

Dilley

Ochitill

Perl

Volberding

(1985). Findings in psychiatric consultations with patients with acquired immune deficiency syndrome. American Journal of Psychiatry, 142, 82–86.

17.

DiMatteo

(2004). Social support and patient adherence to medical treatment: A meta-analysis. Health Psychology, 23, 207–218.

18.

Domínguez

Watkins

(2003). Creating networks for survival and mobility: Social capital among African American and Latin American low-income mothers. Social Problems, 50, 111–135.

19.

Draine

Ahuja

Altice

F. L.

Arriola

K. J.

Avery

A. K.

Beckwith

C. G.

… Tinsley

M. J.

(2011). Strategies to enhance linkages between care for HIV/AIDS in jail and community settings. AIDS Care, 23, 366–377.

20.

Dwyer

Fish

Gallucci

Walker

(2011). HIV care in correctional settings. Retrieved from http://hab.hrsa.gov/deliverhivaidscare/clinicalguide11/cg-105_correctional_settings.html

21.

Facon

Nuchadee

(2010). An item analysis of Raven’s colored progressive matrices among participants with Down syndrome. Research in Developmental Disabilities, 31, 243–249.

22.

Falkin

G. P.

Strauss

S. M.

(2003). Social supporters and drug use enablers: A dilemma for women in recovery. Addictive Behaviors, 28, 141–155.

23.

Fleishman

Sherbourne

Crystal

Collins

Marshall

Kelly

… Hays

(2000). Coping, conflictual social interactions, social support, and mood among HIV-infected persons. American Journal of Community Psychology, 28, 421–453.

24.

Fontaine

Gilchrist-Scott

Denver

Rossman

(2012). A families and reentry: Unpacking how social support matters. Washington, DC: The Urban Institute.

25.

Fontana

Beckerman

(2007). Recently released with HIV/AIDS: Primary care treatment needs and experiences. Journal of Health Care for the Poor and Underserved, 18, 699–714.

26.

Gallant

(2003). The influence of social support on chronic illness self-management: A review and directions for research. Health Education & Behavior, 30, 170–195.

27.

Galvan

Davis

Banks

Bing

(2008). HIV stigma and social support among African Americans. AIDS Patient Care and STDs, 22, 423–436.

28.

Gantschnig

Page

Fisher

(2011). Cross-regional validity of the assessment of motor and process skills for use in Middle Europe. Journal of Rehabilitation Medicine, 44, 151–157.

29.

Giordano

Gifford

White

Almazor

Rabeneck

Hartman

… Morgan

(2007). Retention in care: A challenge to survival with HIV infection. Clinical Infectious Diseases, 44, 1493–1499.

30.

Glaze

(2011). Correctional populations in the United States, 2010 Bulletin. Washington, DC: Bureau of Justice Statistics.

31.

Granovetter

(1973). The strength of weak ties. American Journal of Sociology, 78, 1360–1380.

32.

Greenleaf

(1992). Measuring extreme response style. The Public Opinion Quarterly, 56, 328–351.

33.

Hammett

Harmon

Rhodes

(2002). The burden of infectious disease among inmates of and releasees from US correctional facilities, 1997. American Journal of Public Health, 92, 1789–1794.

34.

Hays

Sherbourne

Mazel

(1994). User’s Manual for Medical Outcomes Study (MOS) Core measures of health-related quality of life. Santa Monica, CA: RAND Corporation.

35.

Heaney

Israel

(1996). Social networks and social support. In Heaney

Israel

(Eds.), Health behavior and health education, theory, research, and practice (pp. 179–205). San Francisco, CA: Jossey-Bass.

36.

Huba

Melchior

(1996). Module 46: Social Support Form. The Measurement Group, and HRSA/HAB’s Special Projects of National Significance Cooperative Agreement Steering Committee. Retrieved from http://themeasurementgroup.com/about-the-measurement-group/

37.

Hyman

S. M.

Gold

S. N.

Cott

M. A.

(2003). Forms of social support that moderate PTSD in childhood sexual abuse survivors. Journal of Family Violence, 18, 295–300.

38.

Institute for Criminal Policy Research. (2016). World Prison Brief. Retrieved March 18, 2016, from http://www.prisonstudies.org/country/united-states-america

39.

Kessler

McLeod

(1985). Social support and mental health in community samples. In Syme

S. C. S. L.

(Ed.), Social support and health (pp. 219–240). San Diego, CA: Academic Press.

40.

Kim

Mazza

(2013). Reliability, validity, and item response of MOS social support score among incarcerated women. Women and Criminal Justice, 24, 1–21.

41.

King

Murray

Salomon

Tandon

(2004). Enhancing the validity and cross-cultural comparability of measurement in survey research. American Political Science Review, 98, 191–207.

42.

Knudsen

H. K.

Leukefeld

Havens

J. R.

Duvall

J. L.

Oser

C. B.

Staton-Tindall

… Inciardi

J. A.

(2008). Partner relationships and HIV risk behaviors among women offenders. Journal of Psychoactive Drugs, 40, 471–481.

43.

La Vigne

Shollenberger

Debus

(2009). One year out: Tracking the experiences of male prisoners returning to Houston, Texas. Washington, DC: Urban Institute.

44.

Larios

Davis

Gallo

Heinrich

Talavera

(2008). Concerns about stigma, social support and quality of life in low-income HIV-positive Hispanics. Ethnicity & Disease, 19, 65–70.

45.

Latkin

Donnell

Celentano

Aramrattna

Liu

Vongchak

… Metzger

(2009). Relationships between social norms, social network characteristics, and HIV risk behaviors in Thailand and the United States. Health Psychology, 28, 323–329.

46.

Lee

Peterson

Dixon

(2010). Rasch calibration of physical activity self-efficacy and social support scale for persons with intellectual disabilities. Research in Developmental Disabilities, 31, 903–913.

47.

Lichtenstein

Laska

Clair

(2002). Chronic sorrow in the HIV-positive patient: Issues of race, gender, and social support. AIDS Patient Care and STDs, 16, 27–38.

48.

Lin

(2001). Social capital: A theory of social structure and action. New York, NY: Cambridge University Press.

49.

Lin

Simeone

R. S.

Ensel

W. M.

Kuo

(1979). Social support, stressful life events, and illness: A model and an empirical test. Journal of Health and Social Behavior, 20, 108–119.

50.

Linacre

(2007). A user’s guide to FACETS: Rasch-model computer programs. Retrieved from http://www.winsteps.com/index.htm

51.

Litz

Penk

Gerardi

Keane

(1991). The assessment of posttraumatic stress disorder. In Saigh

(Ed.), Post-traumatic stress disorder: A behavioral approach to assessment and treatment (pp. 109–178). New York, NY: Pergamon Press.

52.

McCoy

Strauss

MacDonald

Leone

Eron

Miller

(2009). Social support and delays seeking care after HIV diagnosis, North Carolina, 2000–2006. AIDS Care, 21, 1148–1156.

53.

McDowell

(2006). Social health. In McDowell

(Ed.), Measuring health: A guide to rating scales and questionnaires (3rd ed., pp. 150–205). Oxford, England: Oxford Press.

54.

McHorney

Tarlov

(1995). Individual-patient monitoring in clinical practice: Are available health status surveys adequate? Quality of Life Research, 4, 293–307.

55.

McPherson

Smith-Lovin

Cook

(2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415–444.

56.

Mears

D. P.

Mancini

Gertz

Bratton

(2008). Sex crimes, children, and pornography: Public views and public policy. Crime & Delinquency, 54, 532–559.

57.

Morenoff

Harding

(2014). Incarceration, prisoner reentry, and communities. Annual Review of Sociology, 40, 411–429.

58.

Morren

Gelissen

Vermunt

(2013). Exploring the response process of culturally differing survey respondents with a response style a sequential mixed methods study. Field Methods, 25, 162–181.

59.

Muraki

(1990). Fitting a polytomous item response model to Likert-type data. Applied Psychological Measurement, 14, 59–71.

60.

Nichols

Speer

Watson

Vergon

Vallee

Meah

(2002). Aging with HIV. San Diego, CA: Elsevier.

61.

Norman

Sloan

Wyrwich

(2003). Interpretation of changes in health-related quality of life: The remarkable universality of half a standard deviation. Medical Care, 41, 582–592.

62.

Nunn

Cornwall

Bazerman

Loewenthal

Beckwith

(2010). Linking HIV-positive jail inmates to treatment, care, and social services after release: Results from a qualitative assessment of the COMPASS Program. Journal of Urban Health, 87, 954–968.

63.

Nurullah

A. S.

(2012). Received and provided social support: A review of current evidence and future directions. American Journal of Health Studies, 27, 173–188.

64.

Offer

(2012). The burden of reciprocity: Processes of exclusion and withdrawal from personal networks among low-income families. Current Sociology, 60, 1–18.

65.

Orrick

Worrall

Morris

Piquero

Bales

Wang

(2011). Testing social support theory: A multilevel analysis of recidivism. Journal of Criminal Justice, 39, 499–508.

66.

Parkerson

Jr Broadhead

Tse

(1995). Perceived family stress as a predictor of health-related outcomes. Archives of Family Medicine, 4, 253–260.

67.

Pratt

Godsey

(2006). Social support, inequality, and homicide: A cross-national test of an integrated theoretical model. Criminology, 41, 611–644.

68.

Quillian

Redd

(2006). Can social capital explain persistent racial poverty gaps? (National Poverty Center Working Paper Series). Ann Arbor, MI: National Poverty Center.

69.

Reckase

(1979). Unifactor latent trait models applied to multifactor tests: Results and implications. Journal of Educational Statistics, 4, 207–230.

70.

Rintamaki

Davis

Skripkauskas

Bennett

Wolf

(2006). Social stigma concerns and HIV medication adherence. AIDS Patient Care STDS, 20, 359–368.

71.

Rose

Clear

(2001). From prison to home: The effect of incarceration and reentry on children, families, and communities incarceration, reentry, and social capital: Social networks in the balance. Washington, DC: Substance Abuse and Mental Health Services Administration (SAMSHA).

72.

Rose

Clear

(2002). Incarceration, reentry, and social capital: Social networks in the balance. Washington, DC: U.S. Department of Health and Human Services.

73.

Ryan

Huebner

Diaz

R. M.

Sanchez

(2009). Family rejection as a predictor of negative health outcomes in white and Latino lesbian, gay, and bisexual young adults. Pediatrics, 123, 346–352.

74.

Sabol

Couture

(2008). Prison inmates at midyear 2007. Washington, DC: Department of Justice.

75.

Sampson

R. J.

(2004). Neighborhood and community: Collective efficacy and community safety. New Economy, 11, 106–113.

76.

Sampson

R. J.

Morenoff

J. D.

Smedley

B. D.

Syme

S. L.

(2000). Public health and safety in context: Lessons from community-level theory on social capital promoting health: Intervention strategies from social and behavioral research. Washington, DC: National Academy Press, Institute of Medicine.

77.

Schuman

(2008). Method and meaning in polls and surveys. Cambridge, MA: Harvard University Press.

78.

Shahriar

Delate

Hays

Coons

(2003). Commentary on using the SF-36 or MOS-HIV in studies of persons with HIV disease. Health and Quality of Life Outcomes, 9, 1–7.

79.

Sherbourne

(1988). The role of social support and life stress events in use of mental health services. Social Science & Medicine, 27, 1393–1400.

80.

Sherbourne

Meredith

Rogers

Ware

Jr (1992). Social support and stressful life events: Age differences in their effects on health-related quality of life among the chronically ill. Quality of Life Research, 1, 235–246.

81.

Sherbourne

Stewart

(1991). The MOS social support survey. Social Science & Medicine, 32, 705–714.

82.

Shinn

Lehmann

Wong

N. W.

(1984). Social interaction and social support. Journal of Social Issues, 40, 55–76.

83.

Simoni

Frick

Huang

(2006). A longitudinal evaluation of a social support model of medication adherence among HIV-Positive men and women on antiretroviral therapy. Health Psychology, 25, 74–81.

84.

Small

M. L.

(2009). Unanticipated gains. New York, NY: Oxford University Press.

85.

Smith

S. S.

(2000). Mobilizing social resources: Race, ethnic, and gender differences in social capital and persisting wage inequalities. The Sociological Quarterly, 41, 509–537.

86.

Spaulding

Seals

Page

Brzozowski

Rhodes

Hammett

(2009). HIV/AIDS among inmates of and releasees from US correctional facilities, 2006: declining share of epidemic but persistent public health opportunity. PLoS One, 4, 1–8. doi:10.1371/journal.pone.0007558.

87.

Staton-Tindall

Royse

Leukefeld

(2007). Substance use criminality, and social support: An exploratory analysis with incarcerated women. American Journal of Drug and Alcohol Abuse, 33, 237–243.

88.

Stein

M. S.

Spaulding

Cunningham

Messina

Kim

Chung

… Flanigan

(2012). HIV-positive and in jail: race, risk factors, and prior access to care. AIDS and Behavior, 17, S108–117.

89.

Thissen

Steinberg

(1986). A taxonomy of item response models. Psychometrika, 51, 567–577.

90.

Thoits

(1982). Conceptual, methodological, and theoretical problems in studying social support as a buffer against life stress. Journal of Health and Social Behavior, 23, 145–159.

91.

Thomas

Sampson

(2005). High rates of incarceration as a social force associated with community rates of sexually transmitted infection. Journal of Infectious Diseases, 191, S55–S60.

92.

Tross

Hirsch

(1988). Psychological distress and neuropsychological complications of HIV infection and AIDS. American Psychologist Special Issue: Psychology and AIDS, 43, 929–934.

93.

Uchino

B. N.

(2006). Social support and health: A review of physiological processes potentially underlying links to disease outcomes. Journal of Behavioral Medicine, 29, 377–387.

94.

Van Vaerenbergh

Thomas

(2013). Response styles in survey research: A literature review of antecedents, consequences, and remedies. International Journal of Public Opinion Research, 25, 195–217.

95.

Victor

Bowling

Bond

Scambler

(2003). Loneliness, social isolation and living alone in later life. London, England: ESRC Growing Older Programme.

96.

Wakeman

Rich

(2010). HIV Treatment in US Prisons. HIV Therapy, 4, 505–510.

97.

Wellman

Wortley

(1990). Different strokes from different folks: Community ties and social support. American Journal of Sociology, 96, 558–588.

98.

Williams

Kim

Meyer

Spaulding

Teixeira

Avery

… Ouellet

(2013). Gender differences in baseline health, needs at release, and predictors of care engagement among HIV positive clients leaving jail. AIDS and Behavior, 17, S195–S202.

99.

Wilson

(2005). Constructing measures: An item response modeling approach. Mahwah, NJ: Lawrence Erlbaum.

100.

Wohl

Galvan

Myers

Garland

George

Witt

… Carpio

(2011). Do social support, stress, disclosure and stigma influence retention in HIV care for Latino and African American men who have sex with men and women? AIDS and Behavior, 15, 1098–1110.