PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Med Care. Author manuscript; available in PMC 2013 June 20.
Published in final edited form as:
PMCID: PMC3687080
NIHMSID: NIHMS319737

Gender Differences in Multiple Underlying Dimensions of Health-Related Quality of Life are Associated with Sociodemographic and Socioeconomic Status

Abstract

Purpose

To examine whether gender differences in summary health-related quality-of-life (HRQoL) are due to differences in specific dimensions of health, and whether they are explained by sociodemographic and socioeconomic (SES) variation.

Methods

National Health Measurement Study collected cross-sectional data on a national sample of 3648 black and white non-institutionalized adults ages 35–89, including the SF-36v2™ Mental (MCS) and Physical Component Summary (PCS) scores and HRQoL indexes—SF-6D, EQ-5D, HUI2/3 and QWB-SA. Structural equation models were used to explore gender differences in physical, psychosocial, and pain latent dimensions of the five indexes, adjusting for sociodemographic and SES indicators. Observed MCS and PCS scores were examined in regression models to judge robustness of latent results.

Results

Men had better estimated physical and psychosocial health and less pain than women with similar trends on SF-36v2™ scores. Adjustments for marital status or income reduced gender differences more than other indicators. Partial-factorial-invariance of HRQoL attributes adjusted-results supported the presence of gender differentials, but also indicated that these differences are impacted by dimensions being related to some HRQoL attributes differently by gender.

Conclusions

Men have better estimated health on three latent dimensions of HRQoL—physical, psychosocial, and pain—comparable to gender differences on the observed MCS and PCS scores. Gender differences are partly explained by sociodemographic and SES factors, highlighting the role of socioeconomic inequalities in perpetuating gender differences in health outcomes across multiple domains. These results also emphasize the importance of accounting for measurement invariance for meaningful comparison of group differences in estimated means of self-reported measures of health.

Keywords: gender differences, men’s health, women’s health, health-related quality-of-life, patient-reported outcomes, factor analysis, structural equation modeling, health dimensions, EQ-5D, HUI2, HUI3, SF-6D, QWB-SA

INTRODUCTION

Gender differences are well documented for multiple health-related outcomes, such as disease prevalence, mortality, health behaviors and care utilization[14]. These differences are not all in the same direction, since women tend to self-report worse health and utilize more health care services, but live longer than men. The increasing life expectancy (LE) for both men and women with a narrowing gender gap across the past 30 years[5] highlights the importance of measuring and optimizing health-related quality of life (HRQoL) for both genders in the United States (US).

Preference-based measures of HRQoL collect self-reports of function and well-being on multiple health-related attributes and combine these using preference weighted algorithms into a single summary score anchored by 0 (dead) and 1 (full health). These summary measures of overall health are increasingly used to examine outcomes of clinical trials and interventions, to study population trends in health and health disparities, and to assess the cost-effectiveness of new medical technologies. Five commonly-used indexes of HRQoL in the US include the Short Form 6 dimension (SF-6D)[6], EuroQol 5 dimension (EQ-5D)[7], the Health Utilities Index Mark 2 (HUI2)[8] and Mark 3 (HUI3)[9], and the Quality of Well-Being Scale Self-Administered form (QWB-SA)[10]. Prior research indicates that while women have worse self-reported HRQoL across all five of these indexes, small gender differences remain across most of these indexes after sociodemographic and socioeconomic (SES) variation between men and women is taken into account[11]. It is unknown whether gender differences in HRQoL are driven by specific underlying dimensions of health, or whether gender differences exist across all underlying health dimensions.

Physical and mental health have been identified as underlying domains of generic health measures[12], and the SF-36v2™ yields separate Mental (MCS) and Physical Component Summary (PCS) scores[13]. Factor analysis has suggested that the five HRQoL indexes share three underlying health dimensions: physical, psychosocial, and pain[14]. Determining whether gender differences are limited to or stronger in certain dimensions will generate understanding of the source of gender disparities on the overall aggregate summary level of HRQoL.

The present paper is the third in a sequence of papers using the National Health Measurement Study (NHMS) data to examine gender differences in HRQoL. The first[11] identified gender differences measured by the five HRQoL indexes. The second[14] decomposed those indexes into latent attributes of health common to the five measures. The present paper asks whether summary HRQoL gender differences stem from differences in one or more of the underlying latent dimensions of health. We first examine whether there are gender differences in how the attributes capture the latent dimensions, and then test the hypothesis: women report lower health on underlying physical, psychosocial and pain dimensions. The existence of SES disparities in HRQoL[16] and the importance of sociodemographic and SES factors in explaining gender differences in summary measures of HRQoL[11] lead us to test a second hypothesis: differences in sociodemographic and SES characteristics partly explain gender differences in the underlying dimensions. Finally, we conduct a parallel analysis of the widely-used mental and physical components—SF-36v2™ MCS and PCS—to judge robustness of gender differences detected in the latent dimensions.

METHODS

Data and Subjects

Data are from the NHMS[15], a national cross-sectional random-digit-dial telephone survey of 3,844 non-institutionalized adults, ages 35 to 89 years, residing in the continental US. Data were collected between June 2005 and August 2006. Four HRQoL questionnaires, the EuroQol 5-Dimension (EQ-5D), the Medical Outcomes Study Short Form 36-Item survey (SF-36v2™), the Health Utilities Index Mark 2 (HUI2) and Mark 3 (HUI3), and the self-administered version of the Quality of Well-being Scale (QWB-SA), were administered in random order to all respondents via a computer-adaptive-telephone-interview and used to estimate the five utility index scores: SF-6D[6], EQ-5D[7], HUI2[8] and HUI3[9] and QWB-SA[10]. The SF-36v2™ questionnaire was also used to compute the MCS and PCS scores, summarizing vitality, social functioning, role limitations due to emotional problems, mental health, physical functioning, role limitations due to physical health problems, bodily pain and general health scales[13]. Survey weights were constructed based on the NHMS sampling design and post-stratified by gender, race and age. This current analysis includes respondents who reported their race/ethnicity as either white or African American/black, resulting in a sample size of 3648.

Measures

Each of the five HRQoL indexes—SF-6D, EQ-5D, HUI2, HUI3 and QWB-SA—produces a summary score based on combining its respective attributes by methods and models specific to the index. Item responses on each HRQoL instrument are used to form attributes. Attributes are health scales focused on a single aspect of health and function (e.g., physical functioning, anxiety/depression) with discrete levels, where each level is defined by a descriptor of the subject’s level of impairment or assessment of their health or well-being. Combination of discrete attribute values defines health states, which include one level from each attribute covered by the index. For the SF-6D, QWB-SA, HUI2 and HUI3, attribute values were individually preference-scored, hence the standard scoring systems for these measures lead to an ordering of the attribute levels. Preference scoring for EQ-5D was done at the overall health state level.

For these analyses, the following attributes were used: six health attributes from SF-6D (with 2–5 levels per attribute), estimated from 11 out of 36 questions on SF-36v2™; five attributes from EQ-5D (3 levels each); four attributes from QWB-SA (3–10 levels); and one and eight attributes from HUI2 and HUI3 (4–6 levels), respectively. Distinct preference scores on each of these attributes were converted to ordinal values for modeling to allow the levels on each attribute to be estimated freely (non-equidistant levels). The five ordinal attributes represented by single items on EQ-5D were used. As a final step in preparing the data, all ordinal attribute variables were recoded such that higher levels indicate better estimated health. HUI2 and HUI3 indexes are scored from the same questionnaire and result in collinear attributes. Hence, only the self-care attribute was included from the HUI2. A detailed description of these data are published elsewhere[11,14,15].

Covariates

The covariates—gender, age, race, marital status, education and income—were coded as follows:

  • Gender: 1=“men”, 0=“women”
  • Age: “35–44”(reference category), “45–54”, “55–64”, “65–74” and “75–89” years old
  • Race/ethnicity: 1=“white”, 0=“African American/black”
  • Marital Status: “widowed/divorced/separated”, “never married”, “married/living with a partner”(reference category)
  • Education: “less than high school”, “high school graduate”, “some post-secondary education”, and “a college degree or higher”(reference category)
  • Household income over the previous year: “less than $20,000”, “$20,000–$34,999”, “$35,000–$74,999”, and “$75,000 or more”(reference category)

Analyses

Cross-loading[17,18] confirmatory factor analysis (CFA) models of the 24 HRQoL attributes (Table 2) were first fit based on the previously published factor structure[14], separately among men (n=1549) and women (n=2099) and both (n=3648). To explore whether meaningful comparisons of estimated group factor means can be made, we conducted an analysis of factorial invariance (loadings and thresholds) by gender[19,20]. This was done by the χ2 test for difference[19] of nested multiple-group CFA (MGA) models to judge whether imposed model restrictions significantly worsened the model fit in two ways: (1) a model with all parameters set equal across gender compared to models with loadings and thresholds of one attribute at-a-time set free across gender; (2) a model with all attribute parameters set free across gender compared to models with loadings and thresholds of one attribute at-a-time set equal across gender.

Table 2
CFA results for the three factor model for full sample and by gender among black and white adults in NHMS.

The gender differences in the underlying dimensions adjusted for sociodemographic and socioeconomic variation were modeled using the full sample in the structural equation modeling[21] (SEM) framework using the above-described cross-loading factor structure. Five SEM models were fit by simultaneously regressing the three hypothesized latent dimensions on age, gender and race (model 1) followed by one-at-a-time adjustment for the sociodemographic and SES variables: marital status (model 2), education (model 3), and income (model 4). A final model was fit with all covariates in the model simultaneously (model 5). Subsequently, the five models were refit using Multiple-group SEM (MG-SEM) which allowed attributes with significantly different item parameters (loadings and thresholds) across gender (detected in MGA) to be freely estimated while constraining all other model parameters to equality across gender. In MG-SEM, group differences in latent dimensions are captured by latent factor intercepts, hence all covariates were centered to either weighted sample (for age and race/ethnicity since the data is post-stratified to Census 2000) or Census 2000 (marital status, education, income) proportions. Thus, the MG-SEM estimated gender differences in latent dimensions are adjusted for partial factorial invariance.

For sensitivity analysis, we examined whether additional direct regression paths from covariates to HRQoL attributes should be included in the SEM models. Modification indices indicate whether freeing additional regression paths within an SEM model will statistically significantly or meaningfully improve the model fit in terms of χ2 with one degree of freedom and expected parameter change [21].

The CFA and SEM models were estimated using delta parameterization and Mean-adjusted Weighted Least Squares robust estimation geared for modeling ordinal data while using survey weights[2126]. All factor loadings and threshold parameters were allowed to be estimated while factor variances and means were fixed, respectively, at one and zero[21].

The model fit of the CFA and SEM models were evaluated by commonly-used goodness-of-fit statistics: Tucker-Lewis Index (TLI), Comparative Fit Index (CFI) and Root Mean Square Error of Approximation (RMSEA)[21]. TLI and CFI values>.95 and RMSEA<.06 indicate a relatively good fit between the hypothesized model and the observed data[27]. Additionally, differences in CFI (ΔCFI) have been reported relative to a cutoff value of ΔCFI=.002, in investigations of measurement invariance, since χ2 tests are sensitive to sample size and may detect trivial differences in the properties of a measure across groups[28].

To evaluate robustness of the latent results, we fit the five covariate-adjusted models in the weighted-least squares (WLS) regression framework using the observed MCS and PCS scores as outcome variables. As in the SEM analyses, all covariates were modeled as indicator variables and survey weights were used to account for the NHMS survey design.

Descriptive statistics and WLS models were estimated by the SAS/STAT® System for Windows version 9.1 (Copyright 2002–2003 SAS Institute Inc.). The latent model analyses were performed by Mplus Version 5.2[19].

RESULTS

Respondent Characteristics

There were more women than men both in the sample and underlying population (54%) while the 10-year age group and race/ethnicity proportions were similar by gender (Table 1). As in the US census, more women (24% vs. 13%) were widowed, divorced or separated and more men (81% vs. 69%) were married or living with a partner. Fewer women than men had a college degree or higher (38% vs. 45%), and a greater proportion of women (29% vs. 23%) were in one of the two lower income categories (<$20,000; $20,000–$34,999).

Table 1
Sample characteristics among black and white adults in the NHMS.

These weighted proportions of sample characteristics were previously compared to other national surveys, reflecting similarly and well the corresponding US Census 2000 target population by gender, age, race, and marital status[11]. The NHMS had a greater proportion of highest levels of income and education than the general US population. However, all surveys showed more women than men being unmarried/not living with a partner and in the lowest income categories and fewer women with a college degree or higher

CFA Model Results

Table 2 shows the loadings for the hypothesized CFA model, fit by gender and for the full sample, of HRQoL attributes that jointly share the physical, psychosocial, and pain latent dimensions. The CFA loadings and model fit statistics indicating good model fit (CFI, TLI>0.95, RMSEA<0.06) were similar for men and women.

MGA analyses of factorial invariance (not shown) revealed that twelve attributes have statistically significantly (P<.05) different loadings and thresholds across gender—SF-6D physical functioning, social functioning, vitality, role limitation; QWB-SA self-care/usual activities, physical activity, acute/chronic symptoms; HUI2 self-care; HUI3 ambulation, hearing, vision, dexterity—with most of these cross-loading across the dimensions. Partial factor invariance of HRQoL attributes across gender indicates that group differences in estimated factor means may be affected (and possibly inflated) by differences in interpreting certain items and their levels[20]. However, the ΔCFI≤.002 for all χ2 difference tests indicated that subtle differences in model fit were detected.

SEM and WLS Model Results

SEM models were fit in the full sample since at least partial factorial invariance was met and small ΔCFI values indicated possibly subtle differences in model fit[20, 28]. The latent gender effects from the SEM models are summarized in Table 3 and plotted in Figure 1.

Figure 1
Estimated gender differences in three latent dimensions of HRQoL attributes and SF-36v2™ mental and physical component summary scores (PCS; MCS) by adjustment model with women as the reference group. Adjusted-‘dimension’ results ...
Table 3
Structural equation model results of gender differences in latent HRQoL dimensions among black and white adults in NHMS.

The SEM models 1–5 fit the data well (CFI, TLI>.95, RMSEA<.06). Gender differences on latent dimensions were positive, indicating that on average men have better estimated latent physical and psychosocial health and less latent pain than women across models. The trends in gender differences across adjustment models were similar for the three dimensions, although the magnitude of differences was larger on physical and psychosocial dimensions than on the pain dimension. On average, men were 0.18 standard deviation units healthier than women on the physical dimension, 0.18 standard deviation units healthier on the psychosocial dimension and 0.12 standard deviation units healthier on the pain dimension, adjusted for age and race (Table 3, Figure 1). Gender differences decreased across all dimensions of HRQoL when adjusted separately for marital status, education, income and all covariates simultaneously. On the physical and psychosocial dimensions, gender differences were statistically significant (P<.05) despite adjustment, except for a weakly significant (P<.08) result in the simultaneously-adjusted model 5 on the psychosocial dimension. Gender differences on the pain dimension were only statistically significant in model 1 (P<.05; age and race). Partial invariance adjusted MG-SEM results (Figure 1; Table 2, Appendix) show the presence of reduced gender differences on all three latent dimensions with comparable covariate-adjusted trajectory of change to the SEM results described above. Similar findings for gender differences emerged using the observed SF-36v2™ PCS and MCS scores, although as in MG-SEM results were rarely significant (Table 4; Table 2, Appendix).

Table 4
WLS regression results of gender differences in observed SF-36v2™ PCS and MCS among black and white adults in NHMS.

Figure 1 shows the small but consistently positive gender differences (favoring men) with similar trends when adjusted for covariates across latent (physical, psychosocial, pain) and observed (PCS, MCS) variables.

Sensitivity analysis revealed nine significant direct paths for certain covariates (race, gender, age dummies) in the SEM models but no significant results in the MG-SEM models. However, modification indices were small (χ2<25) and inclusion of these direct paths in SEM models did not change estimated gender differences, indicating that freeing these paths may unnecessarily complicate the models. Hence, these direct paths were not included in the final SEM models.

DISCUSSION

We found that previously described gender differences in overall HRQoL are not driven by women’s disadvantage in only one or two underlying dimensions of health. Our results show gender differences on all three dimensions underlying the attributes of five commonly-used HRQoL indexes—SF-6D, EQ-5D, HUI2, HUI3 and QWB-SA. Gender differences, although small, persistently indicate lower estimated health among women than men on physical, psychosocial and pain dimensions and on mental and physical component summary scores of the SF-36v2™. Overall, the aggregate gender differences previously found in HRQoL[11] result about equally from all three underlying dimensions (physical, psychosocial, and pain).

The magnitude of gender differences with and without adjustment for sociodemographic and SES factors varies by latent dimension, with the smallest differences found on pain and the largest on the physical and psychosocial dimensions. Differences are reduced by adjustment for sociodemographic and SES factors. Gender-related variation in income and marital status explain more of the differentials than do age, race and education and the differentials change in parallel for all three dimensions with adjustment (Figure 1). The pattern in estimates in the physical and psychosocial dimensions also parallels that of SF-36v2™ and SF-12v2™ (not shown) PCS and MCS scores. The importance of marital status and income in HRQoL[11,16] have been previously documented, and gender differentials in these characteristics exist in the US population. A 2005 Census report[43] documented more men than women have at least a Bachelor’s degree and men have higher median earnings than women by race and Hispanic origin. Median income of families maintained by women with no husband present was lower than that of other type of families, and in general women 18 and older were more likely than their male counterparts to live in poverty. Hence, our results reinforce veridical impact of women’s social and economic disadvantage on their well-being and health. However, socioeconomic status is difficult to fully capture, since it is defined by multiple factors, and typically just a few indicators are collected in population studies[37,38]. Better measures of social and economic factors, capturing change over the life course, may have resulted in stronger effect of adjustment on gender differences.

Analyses adjusting for partial factorial invariance of HRQoL attributes supported the presence of gender differences on all three latent dimensions, but also indicated that gender differences are impacted by latent dimensions being related to some HRQoL attributes in a different way for men and women. These results stress the importance of examining and accounting for measurement invariance (e.g., differential item functioning) for meaningful comparison of group differences in estimated means of self-reported measures of health.

The relatively smaller gender difference found on the pain dimension, than on physical and psychosocial dimensions, may be partly due to the close association of the pain and physical dimensions. This may have led the physical dimension to account for most of the observed gender difference in the simultaneously-modeled underlying factor structure of HRQoL attributes[14]. Indeed, constructs of physical function and pain have been combined into common attributes by others[10]. For example, some generic indexes of self-reported HRQoL and health status, such as the QWB-SA, summarize questions pertaining to physical health and pain into other attributes (e.g., QWB-SA acute and chronic symptoms)[10].

HRQoL indexes are meant to summarize the common impact of diverse health conditions. Hence, our results do not necessarily imply that specific differences in disease occurrence between men and women are responsible for the observed gender differential or directly associated with sociodemographic and SES differences. However, gender differences found on the latent dimensions are consistent with other studies on women’s health[1,34,3538]. Women experience decreased physical functioning and more pain, possibly because they experience higher prevalence of chronic disabling autoimmune and rheumatologic disorders[1,2931], anemia, thyroid conditions, urinary tract infections, gallbladder conditions, migraines and eczema[1,2933]. Women are more commonly affected with depressive and anxiety disorders than men[1,34] and this may explain our finding that women had lower scores on the underlying psychosocial dimension. Although men are more often affected with other mental health problems, such as antisocial behavior, substance abuse and suicide[1,3], these aspects of health are not well captured by attributes loading on the psychosocial dimension. Additionally, women may have a lower threshold and tolerance for pain, seek treatment for pain more often than men, and give higher pain ratings in experimentally administered stimuli[32,33].

This study has some limitations. We examined gender differences in three dimensions jointly defined by attributes of the five HRQoL indexes: EQ-5D, SF-6D, HUI2, HUI3 and QWB-SA. Although these indexes are commonly-used in the US, it has been suggested that other important dimensions of health may exist. For example, the World Health Organization measure of HRQoL, WHOQOL-BREF[39,40], defines attributes of social relationships and environment. Additionally, subsequent research on gender and health should address subgroups not represented by this study, such as ages beyond 35 to 89, institutionalized people, and those falling into the ”other” racially/ethnically diverse subgroups of the US population. We did not explore how other self-reported measures of health (e.g., symptoms/conditions, disease-targeted measures) relate to gender differences in underlying dimensions of HRQoL. Importantly, we also did not examine how gender differences vary with age. Men tend to die younger than women, possibly leading to healthier men surviving to older age, with a resulting larger gender difference in HRQoL. The life expectancy advantage for women may be outweighed by lower quality of life[41,42]. Additionally, the possibility that gender variations in health behaviors may have a significant impact in explaining gender differences in HRQoL outcomes and mortality remains largely unexplored. We have also considered that differential item functioning may exist by other important subgroups.

The primary strength of this study is the use of a large dataset from a recently surveyed national sample of US adults on five simultaneously administered and commonly-used preference-based indexes of HRQoL. This study presents a first look at the relationship of gender and three underlying health-related dimensions (physical, psychosocial, and pain), jointly defined by attributes of five commonly-used indexes of HRQoL.

CONCLUSION

US men have better estimated health than do women on latent physical, psychosocial and pain dimensions, as well as on the SF-36v2™ physical and mental component summary scores. Gender differences are reduced but not fully explained by adjustments for sociodemographic and SES variation between men and women, except for the pain dimension. All five health-related outcomes showed similar patterns in gender differences when adjusted for sociodemographic and SES variables, with income and marital status contributing the most to explaining gender differences. Our study complements previous findings of gender differences as captured by the five HRQoL indexes, highlights the importance of studying the relationship of gender and health among independent health-related factors of HRQoL, and continues to emphasize the impact of socioeconomic disparities on the well-being of women.

Supplementary Material

Acknowledgment

Support: This research was supported by a grant (P01-AG020679) to the University of Wisconsin, Madison, from the National Institute on Aging and a grant (T32 HS000046) to University of California, Los Angeles, and RAND Corporation, Santa Monica, from the Agency for Healthcare Research and Quality. Ron D. Hays was supported in part by the UCLA Resource Center for Minority Aging Research/Center for Health Improvement in Minority Elderly RCMAR/CHIME), NIH/NIA Grant Award Number P30AG021684, the UCLA/ Drew Project EXPORT, NCMHD, 2P20MD000182, and the UCLA Older Americans Independence Center, NIH/NIA Grant P30-AG028748.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Presented: at the 16th Annual NRSA Trainees Research Conference (June 26, 2010), Boston, Massachusetts.

REFERENCES

1. Bird DC, Rieker PP. Gender and health: The effects of constrained choices and social policies. New York: Cambridge University Press; 2008.
2. Xu JQ, Kochanek KD, Murphy SL, Tejada-Vera B. National vital statistics reports. no 19. vol 58. Hyattsville, MD: National Center for Health Statistics; 2010. Deaths: Final data for 2007.
3. Courtenay WH, McCreary DR, Merighi JR. Gender and ethnic differences in health beliefs and behaviors. Journal of Health Psychology. 2002;7(3):219–231. [PubMed]
4. Bertakis KD, Azari R, Helms LJ, Callahan EJ, Robbins JA. Gender differences in the utilization of health care services. J Fam Pract. 2000;49(2):147–152. [PubMed]
5. Arias E. National vital statistics reports. no 21. vol 58. Hyattsville, MD: National Center for Health Statistics; 2010. United States life tables, 2006.
6. Brazier J, Roberts J, Deverill M. The estimation of a preference-based measure of health from the SF-36. J Health Econ. 2002;21:271–292. [PubMed]
7. Shaw JW, Johnson JA, Coons SJ. US valuation of the EQ-5D health states: development and testing of the DI valuation model. Medical Care. 2005;43(3):203–220. [PubMed]
8. Torrance WG, Feeny DH, Furlong WJ, Barr RD, Zhang Y, Wang Q. Multiattribute utility function for a comprehensive health status classification system: health utilities index mark 2. Medical Care. 1996;34(7):702–722. [PubMed]
9. Feeny D, Furlong W, Torrance GW, et al. Multiattribute and singleattribute utility functions for the health utilities index mark 3 system. Medical Care. 2002;40:113–128. [PubMed]
10. Andresen EM, Rothenberg BM, Kaplan RM. Performance of a self-administered mailed version of the Quality of Well-Being (QWB-SA) questionnaire among older adults. Medical Care. 1998;36:1349–1360. [PubMed]
11. Cherepanov D, Palta M, Fryback DG. Gender differences in health-related quality-of-life are partly explained by sociodemographic and socioeconomic variation between adult men and women in the US: evidence from four US nationally representative data sets. Quality of Life Research. 2010;19(8):1115–1124. [PMC free article] [PubMed]
12. Cella DF, Chang CH, Wright BD, et al. Defining Higher Order Dimensions of Self-Reported Health: Further Evidence for a Two-Dimensional Structure. Eval Health Prof. 2005;28:122. 2005. [PubMed]
13. Ware JE, Jr, Kosinski M, Keller SD. SF-36 Physical and Mental Health Summary Scales: A user’s manual. Boston: Health Institute, New England Medical Center; 1994.
14. Cherepanov D, Palta M, Fryback DG. Underlying Dimensions of the Five Health-Related Quality-of-Life Measures Used in Utility Assessment: Evidence from the National Health Measurement Study. Medical Care. 2010;48(8):718–725. [PMC free article] [PubMed]
15. Fryback DG, Dunham NC, Palta M, et al. US norms for 6 generic health-related quality of life indexes from the National Health Measurement Study. Med care. 2007;45:1162–1170. [PMC free article] [PubMed]
16. Robert SA, Cherepanov D, Palta M, Dunham NC, Feeny D, Fryback DG. Socioeconomic status and age variations in health-related quality of life: Results from the national health measurement study. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences. 2009;64B(3):378–389. [PMC free article] [PubMed]
17. Marsh HW, Muthén B, Asparouhov A, et al. Exploratory structural equation modeling, integrating CFA and EFA: application to students’ evaluations of university teaching. Structural Equation Modeling. 2009;16:439–476.
18. Asparouhov T, Muthén B. Exploratory structural equation modeling. Structural Equation Modeling. 2009;16:397–398.
19. Muthén LK, Muthén BO. Mplus User’s Guide. Fifth Edition. Los Angeles, CA: Muthén & Muthén; 1998–2008.
20. Gregorich SE. Do Self-Report Instruments Allow Meaningful Comparisons Across Diverse Population Groups? Testing Measurement Invariance Using the Confirmatory Factor Analysis Framework. Medical Care. 2006;44(11S3):S78–S94. [PMC free article] [PubMed]
21. Kaplan D. Structural Equation Modeling: Foundations and Extensions. 1st ed. Thousand Oaks, CA: Sage Publications; 1995. pp. 40–129.
22. Asparouhov T. Sampling weights in latent variable model. Structural Equation Modeling. 2005;12:411–434.
23. Muthén B, Asparouhov T. Latent Variable Analysis With Categorical Outcomes: Multiple-Group and Growth Modeling in Mplus. [Accessed October 6, 2010];Mplus Web Notes. No. 4, Version 5. 2002 Available at: http://www.statmodel.com/download/webnotes/CatMGLong.pdf.
24. Dumenci L, Achenbach TM. Effects of estimation methods on making trait-level inferences from ordered categorical items for assessing psychopathology. Psychol Assess. 2008;20:55–62. [PubMed]
25. Flora DB, Curran PJ. An empirical evaluation of alternative methods of estimation for confirmatory factor analysis with ordinal data. Psychol Methods. 2004;9:466–491. [PMC free article] [PubMed]
26. Muthén BO. Mplus Technical Appendices. Los Angeles, CA: Muthén & Muthén; 2006. [Accessed October 6, 2010]. http://www.statmodel.com/download/techappen.pdf.
27. Hu L, Bentler PM. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling. 1999;6(1):1–55.
28. Meade AW, Johnson EC. Power and Sensitivity of Alternative Fit Indices in Tests of Measurement Invariance. Journal of Applied Psychology. 2008;93(3):568–592. [PubMed]
29. Wingard DL. The sex differential in morbidity, mortality, and lifestyle. Ann Rev Pub Health. 1984;5:433–458. [PubMed]
30. Verbrugge LM, Wingard DL. Sex differentials in health and mortality. Women and Health. 1987;12:103–145. [PubMed]
31. Wingard DL, Cohn BA. Variations in disease-specific sex morbidity and mortality ratios in the United States. In: Ory MG, Warner HR, editors. Gender, health, and longevity: Multidisciplinary perspectives. New York, NY: Springer Publishing Co.; 1990. pp. 25–37.
32. Berkley KJ. Sex differences in pain. Behavioral and Brain Sciences. 1997;20(3):371–380. [PubMed]
33. Wilson JF. The Pain Divide between Men and Women. Annals of Internal Medicine. 2006;144(6):461–464. [PubMed]
34. Nolen-Hoeksema S. Gender Differences in Depression. Current Directions in Psychological Science. 2001;10:173.
35. Dentona M, Prusb S, Waltersc V. Gender differences in health: a Canadian study of the psychosocial, structural and behavioural determinants of health. Social Science & Medicine. 2004;58:2585–2600. [PubMed]
36. Rieker PP, Bird DC. Rethinking Gender Differences in Health: Why We Need to Intergrate Social and Biological Perspectives. The Journals of Gerontology: Series B: Psychological Sciences and Social Sciences. 2005;60B:40–48. [PubMed]
37. Adler NE, Stewart J, et al. Reaching for a Healthier Life: Facts on Socioeconomic Status and Health in the U.S. The John D. and Catherine T. Macarthur Foundation Research Network on Socioeconomic Status and Health. 2007
38. Adler NE, Stewart J. Preface to The Biology of Disadvantage: Socioeconomic Status and Health. Ann. N.Y. Acad. Sci. Issue: The Biology of Disadvantage. 2010;1186:1–4. [PubMed]
39. World Health Organization (1993) WHO; 1993. WHOQoL Study Protocol. (MNH7PSF/93.9)
40. Murphy B, Herrman H, Hawthorne G, Pinzone T, Evert H. Australian WHOQoL instruments: User’s manual and interpretation guide. Melbourne, Australia: Australian WHOQoL Field Study Centre; 2000.
41. Kaplan RM, Erickson P. Gender Differences in Quality-Adjusted Survival Using a Health-Utilities Index. American Journal of Preventive Medicine. 2000;18(1):77–86. [PubMed]
42. Kaplan RM, Anderson JP, Ake CF. Gender Differences in Quality-Adjusted Life Expectancy: Results From the National Health Interview Survey. Clinical Journal of Women’s Health. 2001;1(4):191–198.
43. Renee E, Spraggins We the People: Women and Men in the United States. Census 2000 Special Reports. CENSR-20. Issued January 2005. Access at: http://www.census.gov/prod/2005pubs/censr-20.pdf.