A longitudinal model of the dynamics between HMOS' consumer-friendliness and preventive health care utilization.

Xiao, Qian ; Savage, Grant T. ; Zhuang, Weiling 等

INTRODUCTION

Health maintenance organizations (HMOs) are a well-established response to years of escalating costs in the health care system of the United States. This type of group health insurance plan is considered one of most effective tools for managing health care services and containing consumer costs. In this sense, the careful examination of HMOs' performance is of paramount significance. HMOs typically lower consumer cost sharing by requiring only modest co-payments rather than the high deductibles and co-insurance which are more common in other types of group health insurance. Other means by HMOs to manage the range of service offerings and contain care costs include: (a) selecting a network of providers and predetermining the range of health care services to its enrollees; (b) relying on primary care physicians to be care managers who must approve referrals to specialists; (c) using capitation or other financial incentives to encourage cost-effective care; and (d) employing a variety of utilization management tools, such as profiling service use, requiring prior authorization, and using case management for high-cost services.

However, the research evidence to justify HMOs' performance in terms of improving access to preventive care and other health care services is thus far mixed and inconclusive. Results differ as to whether HMOs increase or decrease health care utilization, and whether this relationship has evolved over time (Rizzo, 2005). Reviews of the literature by Robinson and Steiner (1998) and Miller and Luft (1997, 2002) found no conclusive evidence in one direction or the other on the relationship between managed care and quality of care. Studies that focused specifically on the relationship between HMOs and health care utilization found little difference in health outcomes (Rizzo, 2005). As an effort to resolve the problem of the mixed or even conflicting research findings, Xiao and Savage (2008) proposed the multidimensionality aspect of HMOs.

Specifically, they argued that the incongruent conclusions about HMOs' effectiveness may reveal the complication of heterogeneous models of managed care, and that the mixed results could be due to the aggregation of some forms of HMOs that are very effective with other forms that are relatively ineffective. Therefore, rather than asking the questions like "Do HMOs make difference?", or "Are HMOs good for health maintenance?", Xiao and Savage (2008) suggested that the practical research move beyond the simple dichotomy between HMO and non-HMO based care provision, and address the performance in terms of different models of managed care. In their study, recognizing HMOs' variations in the forms and operating mechanisms, Xiao and Savage used consumers' experience with HMOs to distinguish multiform HMOs (Medicaid HMOs and private HMOs in their study) and examined their consumer-friendly characteristics in relationship to consumers' utilization of preventive care services. The 2002 Medical Expenditure Panel Survey (MEPS) was employed to test the hypotheses. The empirical results indicated that HMOs did differ in their consumer-friendly characteristics, and some of these characteristics were significantly associated with utilization of preventive care services.

However, certain limitations in their study such as the cross-sectional research design and the outdated dataset restrict predictability and generalizability, and call for the need of testing the validity and reliability of the research findings. This paper is thus intended to replicate and extend the research by Xiao and Savage (2008) using the most recent available dataset with a national focus.

Replication plays an important role in theory building and justification. It is literally a routinely assumed responsibility in the field of "normal science" (Kuhn, 1962) in order to prove or disprove the emerging theories and models, and is widely recognized as the cornerstones of creating scientific knowledge (Blaug, 1992). Scholars in the field of management have also stressed the need of employing regular replication to verify and elaborate how broadly and in what instances propositions exist and can be used (e.g., Hubbard, Vetter, & Little, 1998; Lindsay, 1994). Each successful replication provides more support for a theory or conception, whereas every failed replication raises further concerns (Popper, 1978). Therefore, the contribution of this study lies in not only testing the validity of the previous research but also in serving as a guide for others interested in replicating health services management models and theories.

Particularly, this study offers significant incremental contributions beyond Xiao and Savage (2008) in several areas. First, this research offers methodology-related advances to comprehensively explore the dynamic interactions between HMOs' consumer-friendly characteristics and consumers' utilization of preventive care services. Xiao and Savage's (2008) work uncovered immediate effects with findings based on a cross-sectional research design, in which the measurement of each variable occurred at a single time period for each case. In contrast, this research introduces a prospective panel design, in which data were collected at distinct periods on the same set of cases and variables in each period. This longitudinal design establishes a temporal order for key variables, and may indicate a predicative relationship and allow for strong causal interpretations. Specifically, our longitudinal equation model takes a dynamic time series approach that not only considers the concurrent relationship involving independent variables (consumers' experience with multiform HMOs) and dependent variables (utilization of preventive services) at different time points to establish a predictive relationship in a temporal order; but also examines accumulative time lagging effects of early experience variables on the later utilization variables over distinct time periods.

Second, the most recent available MEPS panel data set covering two full calendar years of 2006 and 2007 was used to examine whether the proposed temporal relationship between HMOs' consumer-friendly characteristics and the utilization of preventive service holds over time for different sample data. Miller and Luft (1994) pointed out that the performance of managed care organizations differ considerably depending on which local market areas are used for analysis. Consequently, evaluation findings based on data from a small number of plans, providers, or local market areas cannot necessarily be generalized to the nation. They also cautioned that much of the literature relied on relatively old data--data that become less relevant in today's changing health care marketplace. They recommended that future research focus on multiple dimensions of performance and investigate effects on important subgroups. Thus this paper represents a significant step toward addressing Miller and Luft's recommendations and avoiding the limitations of previous studies.

Third, while Xiao and Savage (2008) concluded that HMOs did differ in their consumer-friendly characteristics, and some of these characteristics were significantly associated with utilization of preventive care services in general, this research extends their work by looking into the individual utilization variables within each form of HMO, and substantively examining with ANOVA analysis whether consumers' utilization of preventive services is higher for those HMOs with higher consumer experience ratings than the ones with lower experience ratings.

The rest of the paper is organized as follows. The next section provides the theoretical background followed by the discussion of the conceptual framework. We then describe the sources of data and the model specification. Main findings are presented and discussed next, and the paper ends with concluding remarks, and a discussion of limitations and future research.

THEORETICAL BACKGROUND

Following a similar logic, we drew upon Miller and Luft's (1994) viewpoints to develop our propositions. Miller and Luft (1994) cautioned researchers about drawing conclusions and generalizing from the literature evaluating HMOs. They pointed out that many factors can affect managed care organizations' health care use, expenditure, and quality performance. They divided the most likely factors into three groups: characteristics of the managed care organizations, characteristics of the managed care benefit plans, and characteristics of the markets in which managed care organizations operate. In order to be comparable against Xiao and Savage's (2008) research findings, this study is intended to examine whether the structural characteristics of different HMOs may be associated with the diverse HMO performance in terms of health care utilization.

In this study, HMOs are structurally characterized and distinguished using a 'consumer-centered' approach. With this logic, consumers' experience with HMOs is used to distinguish among multifaceted HMO plans. The choice of a consumer-focus can be warranted for three reasons. First, given the growing interest in accountable care organizations and medical homes as mechanisms for improving care and reducing non-necessary medical care, HMOs present an established way to bundle payments and achieve the goals of these new initiatives. Examining how HMOs perform from a consumer perspective, thus, is of great interest and significance. Second, consumers often have major roles in choosing health care and health plan coverage, which ultimately may have implications for the future viability of different forms of health care delivery and financing. These public perceptions and attitudes also affect the formulation of public policies regarding the regulation and provision of health insurance. Third, the consumer-centered approach, as a market-based solution, highlights consumers' positive roles in the health care by taking into account consumers' values, expectations, and medical needs; and therefore corresponds with the advocacy of consumer-driven health care (Herzlinger, 2004).

Thus this paper contains an in-depth study of Medicaid HMOs and private HMOs' consumer-friendly characteristics that are proposed to serve as the potential explanatory variables for HMOs' diverse performance in terms of consumers' utilization of preventive care services. The study's focus on preventive care utilization is warranted because preventive medicine is the cornerstone of health care. It has important effects on disease progression, morbidity, and mortality. Despite the importance of preventive care, research suggests that such care has been underutilized for many decades in the United States (Schauffler & Rodriguez, 1993). In a report identifying key problems with quality of care in the United States, the President's Advisory Commission on Consumer Protection and Quality in the Health Care Sector ("The State of Health Care Quality", 1998) specifically cited problems with underutilization of preventive care, including flu shots, mammography and screenings for colorectal cancer. While the growth of HMOs was expected to create incentives that would increase preventive care access and utilization, studies of preventive care yield conflicting evidence. Kenkel (1994) found that HMO members used less preventive care as measured by breast examinations and PAP smears. In contrast, Miller and Luft's (2002) review of the literature reported that most studies on preventive care pointed to greater use of preventive care in HMOs. However, the researchers also noted that most of these studies considered cancer screening rather than broad-based comparisons of preventive medicine. Therefore, this study seeks to bridge these gaps in the literature.

Figure 1 depicts a conceptual framework that illustrates the dynamic interrelationship between HMOs' consumer-friendly characteristics and preventive health care utilization. As shown in Figure 1, HMOs' consumer-friendly characteristics influence current and future utilization of preventive care services over distinct time periods. Xiao and Savage's (2008) study employed a cross-sectional approach. The cross-sectional approach is appropriate when variables of interests are time-invariant. However, as HMOs' consumer-friendliness is measured by consumers' evaluations of their experience with HMOs, and the influential valence of experience perception and evaluation varies over time and across different forms of HMOs, thus a dynamic longitudinal approach that explores the cumulative effects and the time-series variations in the effects is necessary to identify the true influence of HMOs' consumer-friendliness.

HMOs' consumer-friendliness influences consumers' preventive care utilization in a way that the distribution of early experience perception and evaluation increases consumers' awareness of or discourages subsequent utilization. Moreover, we also recognize the fleet nature of HMOs' consumer-friendliness that the effects of early experience could gradually wear out if there is no follow-up. Thus utilization behaviors are more likely influenced by the most recent experience than the early ones. This corresponds to a typical feature of a time-series model that observations close together in time will be more closely related than observations further apart. Therefore, rather than considering the effect in a static setting with a focus on the concurrent relationship, this study measures both the concurrent and lagging effects of HMOs' consumer-friendliness over preventive care utilization. We thus propose the following hypotheses:

H1: Medicaid HMOs and private HMOs differ in their consumer-friendly characteristics; and consumers' preventive health care utilization is higher for these HMOs that are more consumer-friendly.

H2: The influence of HMOs' consumer-friendly characteristics on concurrent preventive health care utilization is positive.

H3: The influence of HMOs' consumer-friendly characteristics on preventive health care utilization beyond the concurrent term is positive. However the influence diminishes quickly thus that concurrent term would contain most of the explanatory power.

[FIGURE 1 OMITTED]

Although it is relatively straightforward to postulate that a positive experience enhances utilization and a negative experience reduces it, whether this effect can be unambiguously transformed into actual care utilization is unclear. Indeed, behavioral research has found that experience and attitude does not always predict behavior well (Ajzen & Fishbein, 1980), and that other factors may influence behavior beyond what experience and attitude can explain. Therefore, sociodemographic factors and health status are included as control variables to reduce confounding effects.

Demographic indicators include age, race, education degree, and poverty status. Kenkel (1994) has demonstrated the importance of age and education on preventive medicine, at least with respect to PAP smears and breast examinations. The impact of age involves a trade-off. On the one hand, older age raises the probability of discovering a problem, making preventive screening more beneficial. At the same time, advanced age lessens the potential benefits in terms of increased longevity in the event that screening helps to prevent a medical problem from occurring or worsening. Although the impact of age on the use of preventive care is unclear in terms of whether age will encourage or discourage the use, we deem age is a potential influential factor of preventive care use and should be included as one control variable. Kenkel (1991) also found evidence that better health knowledge explained part of the relationship between schooling and healthy behaviors. Better-educated individuals possess greater knowledge of the benefits associated with preventive care and hence are more proactive in obtaining such care. At the same time, Kenkel (1991) noted that most of the educational effects on healthy behaviors remained even after differences in health knowledge were controlled. Race may relate to preventive medicine because disadvantaged minorities may have less information about the benefits of preventive care, or less generous health insurance plans and / or fewer financial resources generally. Race-specific differences in certain types of disease may also prompt differential use of preventive care. Poverty status relates to disposable financial resources; poor people may tend to underuse preventive care services, resulting in even poorer health status. Thus health status is controlled to avoid the likely understatement of the impact of HMOs on preventive medicine. Evidence suggests that HMO members tend to be younger and healthier (e.g., Glied, 2000; Ellis, 1989), perhaps because sicker individuals are discouraged from joining HMOs, or because such individuals tend to avoid HMOs (Rizzo, 2005). Moreover, health status may affect preventive care to the extent that the providers and / or the patients recognize that the need for screening increases as the patient's health status declines.

METHODOLOGY

Data and Measures

Data for the analysis were obtained from MEPS (Medical Expenditure Panel Survey) 2006 and 2007 full year consolidated data files, and MEPS 2006 and 2007 person round plan public use files. MEPS, cosponsored by the Agency for Healthcare Research and Quality (AHRQ) and the National Center for Health Statistics (NCHS), provides nationally representative estimates of medical treatments and health care expenditures, health status, health insurance coverage, and sociodemographic and economic characteristics for the civilian, noninstitutionalized population in the United States. For details of the sampling procedure, please refer to the corresponding documentation files MEPS HC-111 (October, 2009), MEPS HC-113 (November, 2009), MEPS HC-103 (November, 2008), and MEPS HC-105 (November, 2008). We collected the following information from MEPS 2006 and 2007 full year consolidated data files: consumers' experience with Medicaid HMOs, preventive health care utilization, and control variables such as sociodemographic factors and health status. We collected the information about consumers' experience with private HMOs from MEPS 2006 and 2007 person round plan public use files. Person identifiers were used to integrate four datasets to constitute a panel design study that includes two rounds of interviews covering two full calendar years, and provides data for examining person level changes in our focus variables. A restriction of the MEPS database is that the same set of cases and the variables of interest for this study can only be tracked across the 2006 and 2007 successive calendar years. That said, the survey conducted in prior years contains different and incomparable panel samples, and thus cannot be merged to constitute a full range of time-series data.

Several steps were taken to make the data fit the purpose of analysis. First, cases were deleted if they did not relate to multiple rounds of data collection, and could not contribute to the time-series related analysis. Second, we included in the analysis only adults who were older than 17 years of age and insured by either Medicaid HMOs or private HMOs. Third, outliers were removed when casewise diagnostics showed that some cases were outside three standard deviations; and listwise deletion on key variables of the study was adopted when respondents refused to provide information, the questions were inapplicable to the respondents, or the answers were not ascertained. Thus, the final sample included 619 cases, with 140 insured by Medicaid HMOs and 479 insured by private HMOs. To assess the presence of non-response bias in the data, we compared usable responses after the data-cleaning process against non-usable responses on the major sociodemographic characteristics: age, education, ethnicity and poverty level. The non-response bias test showed that the original sample was well represented by our data in terms of percent distribution of selected characteristics. However, some demographic features were only represented by less than 5 enrollees. For example, for Medicaid HMOs, the sample data only contained 2 American Indians, 4 Asians, and 0 Pacific Islanders, and only 1 person with a bachelor's degree. Therefore, the results of analysis should be interpreted with caution for these variables.

Consumers' experience with HMOs was used to distinguish different levels of consumer-friendliness of multiform HMOs (in this case Medicaid HMOs and private HMOs respectively). Respondents who were insured either by Medicaid HMOs or by private HMOs were interviewed as to their experience with the insurance plans. Question wording was based on an AHRQ-sponsored family of survey instruments designed to measure quality from the consumer's perspective. Selected question items addressed the following topics which represent the major consumer-friendliness characteristics: convenience of getting a personal doctor or nurse, delays in waiting for plan approval for care, helping with finding or understanding plan information, channels for getting help from customer service, experience with paperwork, and general ranking of the plan experience. Answers are quantified using a scale from 1 to 10. Number 1 indicates the worst health plan experience possible featured with the serious issues, and 10 represents the best health plan experience featured with no problems at all. Individual items were added together to constitute consumers' overall rating of their experience with HMOs. In order to simplify our model for a longitudinal, time-series analysis, we used the one-item overall experience rating rather than individual experience items separately to indicate HMOs' consumer-friendliness. This simplification is justified as follows: First, Xiao and Savage reported concerns for the potential of high multi-collinearity among the individual items with all items independently regressed into the equations, as indicated by the correlation coefficients as high as beyond 0.9; second, including all the individual experience items in the temporal model requires listwise deletion on all variables and thus dramatically decreases our sample size and restricts its representation and generalizability; third, using a grand overall experience rating by integrating all individual items together could alleviate confounding effects due to the possibility of the inconsistent relational pattern between each single experience item and preventive care utilization.

Respondents were also asked questions pertaining to whether and how often they had received specific types of preventive medicine. Specifically, the following preventive care services were included in the empirical analysis: blood pressure checks (BPCHEK), cholesterol screenings (CHOLCK), flu vaccination shots (FLUSHT), and routine physical examinations (CHECK). We also included the following preventive care particularly for females: breast exam (BRSTEX), PAP smear test (PAPSMR), and mammography (MAMOGR). Our general assumption is that more preventive care is better care (Rizzo, 2005). It is difficult to argue that receiving more frequent blood pressure checkups, cholesterol screenings, physical examination, or breast examinations, for example, is on average bad. More controversial are mammography screenings, which expose the patient to small amounts of radiation per exam, and which have fairly high false-positive rates (Rizzo, 2005). In other words, the high false-positive rates and the resulting unnecessary radiation exposure along with the additional exam costs could be the possible sources for the contradictory relations between HMOs' consumer-friendliness and use of mammography screenings, if the empirical results contradicted with our hypothesis. For descriptive statistics on the preventive care utilization variables please refer to Appendix A.

Sociodemographic factors and health status were included as control variables. Age was calculated using data source year minus birth year. Education was recoded into four categories: high school and less, bachelor's degree, master's degree, and doctorate degree. Race included 6 categories: white, black, American Indian, Asian, Pacific Islander, and multiple races. A poverty status variable was constructed by dividing family income by the applicable poverty line based on family size and composition, with the resulting percentages grouped into 5 categories: poor (less than 100%), near poor (100% to less than 125%), low income (125% to less than 200%), middle income (200% to less than 400%), and high income (greater than or equal to 400%). Respondents were also asked to indicate their overall health status as excellent, very good, good, fair, or poor.

Model Specification

H1 was examined using one-way ANOVA. Besides comparing Medicaid HMOs and private HMOs in terms of their consumer-friendly characteristics as Xiao and Savage did in their study (2008), we further look into the individual utilization variables within each form of HMO, and substantively evaluate whether consumers' utilization of preventive services is higher for those HMOs with higher consumer-friendliness ratings than the ones with lower ratings.

Recognizing that interactions between HMOs' consumer-friendly characteristics and preventive care utilization can go beyond the concurrent term, we developed a longitudinal equation model that reflects time-varying dynamics and includes the multi-lag terms of HMOs' consumer-friendliness to evaluate H2 and H3. Such a specification helps identify the temporal effects of HMOs' consumer-friendly characteristics on preventive care utilization.

In addition, following extant research, we used a log-linear formulation in our model to deal with the potential issue of range restriction in variable values (e.g., Duan, Gu, & Whinston, 2008; Liu, 2006). The log-linear formulation is consistent with theoretical models of a multistage consumer decision making process, where utilization of preventive care services can be viewed as a series of conditional probabilities applied to the consumer base. (For example, a simple two-stage consumer decision model implies the following: preventive care utilization of consumers with HMOs = consumer base * percentage of consumers enrolled in the HMOs * percentage of consumers using the preventive care given their experience with HMOs.) A log transformation converts the relationship into a linear form for empirical estimation (Duan, Gu, & Whinston, 2008). Moreover, log transformation smoothes the distribution of variables in the linear regression, and the estimated coefficients of the log-linear form directly reflect the elasticity of independent and dependent variables (Duan, Gu, & Whinston, 2008).

Finally, we included control variables as fixed effects in the model to simplify the model specification and capture the non-time varying exogenous factors. Considering control variables are not our research focus, and the data collection is only across two calendar years, it is reasonable to believe that control factors like health status and poverty status do not change dramatically in such a relatively short period of time. Fixed effects estimation allows the error term to be arbitrarily correlated with other explanatory variables, thus making the estimation results more robust (Duan, Gu, & Whinston, 2008).

The equation model is specified as follows:

(1) [log(Preventive care utilization).sub.it] = Constant + [[alpha].sub.0] [log(HMOs' consumer friendliness).sub.it] + [summation].sup.J.sub.j=1] [[alpha].sub.j] [log(HMOs' consumer friendliness);.sub.i,t-1] + [[beta].sub.1] [log(Age).sub.i] + [[beta].sub.2] [log(Race).sub.i] + [[beta].sub.3] [log(Perceived health status).sub.i] + [[beta].sub.4] [log(Poverty status).sub.i] + [[beta].sub.6] [log(Education).sub.i]

For each round of data collection separately (t=1, 2), where i indexes respondents and is omitted hereafter to avoid notation clutter. [(Preventive care utilization).sub.it] [(HMOs' consumerfriendliness).sub.it] denotes the preventive care utilization of respondent i at time t. [(HMOs' consumer friendliness).sub.it] denotes the HMOs' consumer-friendliness evaluation of respondent i at time t. [[summation].sup.J.sub.j=1] [[alpha].sub.j] [log(HMOs' consumer friendliness).sub.i,t-j] expresses the linear addition of multi-lag terms of[ (HMOs' consumer friendliness).sub.it]. [[summation].sup.J.sub.j=1] [[alpha].sub.j] [log(HMOs consumer friendliness).sub.i,t-j] As the MEPS data set only contains two rounds of data collection for our key variables, one lag term is estimated in the analysis.

RESULTS AND DISCUSSIONS

Sample Profile

Table 1 provides a demographic and health status profile of the sample enrolled in Medicaid HMOs and private HMOs. As expected, there were statistically significant differences between Medicaid HMO and private HMO enrollees among education, poverty status, and perceived health status, which supported their inclusion as control variables. Since Medicaid is aimed at low-income Americans, it is understandable that a predominant proportion of Medicaid HMO enrollees were near or under the poverty line, while the private HMO enrollees were characterized by middle- and high- income levels. Correspondingly, Medicaid and private HMO enrollees tended to demonstrate relevant features in terms of their education and health status since, as noted above, education level can influence poverty status; and poverty status could be the reason for better or worse health status. Data distribution for Medicaid HMO and private HMO enrollees suggested that the two samples were quite similar across age and race variables.

RESULTS

As data collection on our variables of interests ran through two calendar years, table 2 provides the correlation matrix with dependent variables and independent variables collected in a temporal order. Similar patterns existed for the variables collected on the concurrent term.

ANOVA results that compare the individual indicators of consumer-friendliness between the two types of HMOs are summarized in table 3. Six out of the seven consumer-friendly variables were significantly different between Medicaid HMOs and private HMOs at the .05 level. The only non-significant item, "paperwork for plan", examines the workload of filling out paperwork for the health plans; both Medicaid HMOs and private HMOs were similar in this aspect. We also conducted a one-way multiple analysis of variance (MANOVA), controlling for the differences in the sample enrolled in each type of insurance by entering six control variables (age, sex, race, education, poverty level, and health status) in the model and obtained similar results. Because of space considerations, we only reported the ANOVA results. In general, we found evidence that different types of HMOs (in this case Medicaid HMOs and private HMOs) can be distinguished by their individual consumer-friendly characteristics.

ANOVA results that compare the overall rating of HMOs' consumer-friendliness and related preventive care utilization between the two types of HMOs are summarized in Table 4. Medicaid HMOs and private HMOs significantly differed in their overall rating of consumer-friendliness at the .05 level. Five out of 7 variables of preventive care utilization matched the distribution that care utilization was significantly higher for HMOs that were more consumer-friendly at the level .05. In general, we found evidence that different types of HMOs (in this case Medicaid HMOs and private HMOs) can be distinguished by their overall rating of consumer-friendly characteristics. Yet ANOVA analysis did not show the mean difference for care utilization of flu shot and blood pressure checkup between the two HMOs. Thus H1 was partially supported.

ANOVA results demonstrated some features consistent with the correlation matrix in Table 2. Table 2 showed a significant positive correlation between poverty status and consumers' evaluation of HMOs' friendliness; that is, people with higher economic status tended to assign a positive rating. Correspondingly, our sample showed that private HMOs had a higher rating of consumer-friendliness, and five out of 7 preventive care utilizations were higher for private HMOs compared with Medicaid HMOs. Yet the results need to be interpreted with caution given that the majority of the sample represents white people older than 45 with an education lower than Bachelor's degree.

The longitudinal model estimation results are reported in Table 5. log(HMOs' consumer-friendliness) it was a significant predictor for five out of seven variables of preventive care utilization (i.e., blood cholesterol check, routine checkup, blood pressure test, breast exam, and mammogram), entailing some support for H2. The time lagged term [log(HMOs' consumer-friendliness).sub.i,t-1] was positively and significantly related to [log(BRSTEX).sub.it] at the .05 level, and positively associated with [log(CHECK).sub.it] and [log(MAMOGR).sub.it] at the .1 level. One obvious feature was that, compared against the concurrent term, the influence magnitude of the time lagged term had significantly diminished in a way that either the effects wore out totally (for example, for blood cholesterol check and blood pressure test, effects of HMOs' consumer-friendliness were significant on the concurrent term, but insignificant on the lagged term), or the effect size diminished over time (for example, for routine checkup and mammogram, effects of HMOs' consumer-friendliness were significant on the concurrent term at the .05 level, but only significant on the lagged term at the .1 level; for breast exam, the effect size of HMOs' consumer-friendliness on the concurrent term was .10 larger than [beta] coefficient .082 on the lagged term). For [log(BRSTEX).sub.it] as [beta] coefficients were both significant at the .05 level for two successive terms of HMOs' consumer-friendliness, we checked for the difference in coefficients between the concurrent term and the lag term. The P difference .028 corresponds to a t-statistic of 7.70 (p < .01). It showed that the decline of influence was statistically significant. Thus H2 was partially supported. Given the log-linear formation, the coefficients of significantly affected preventive care variables suggested the following utilization pattern: For every 10 percent increase in the experience rating of HMOs' consumer-friendliness, the blood cholesterol check increased by 1.01 percent on the concurrent term; the routine checkup increased by 0.85 percent on the concurrent term, and by 0.75 on the one time lag term; the breast exam increased by 1.1 percent on the concurrent term, and by 0.82 on the one time lag term; and the mammogram test increased by 1.04 percent on the concurrent term, and by 0.8 on the one time lag term.

Consistent with Xiao and Savage's findings (2008), obtaining a flu shot was not significantly associated with HMOs' consumer-friendliness on either term. A possible reason could be that flu shot vaccination is becoming a routine care practice. People are well informed about the benefits of flu vaccinations since influenza is one of 10 leading causes of death ("National vital statistics report", 2005). A similar explanation applies to the pap smear test: women have increased their awareness for the need of regular pap smear testing to inspect, treat, and prevent ectocervix-related abnormalities.

As expected, age was positively and significantly associated with five out of seven preventive care utilization variables, demonstrating that increased age generally raises the need for preventive screening to prevent a medical problem from occurring or worsening. Yet Rizzo (2005) noted that the relationship between age and preventive care is unclear a priori. On the one hand, older persons are at greater risk of illness, increasing the returns to directing preventive care at them; and at the same time, physicians are well aware that the need for routine preventive care increases for aging individuals, and are more likely to request preventive health intervention for their older patients. On the other hand, the shorter life expectancy of older individuals limits the benefits of preventive care. Race was not related to any variable of preventive care utilization. This finding could be due to sample representation bias, since the majority of the sample is white; or it could be that in this sample minority and majority populations have equal access to preventive care benefits. This result is similar to those reported by Hass and colleagues (2002) that the benefits of managed care expected to be associated with the greater use of some preventive care are not apparent for black persons or Asian/Pacific Islanders enrolled in HMOs. Education significantly affected the use of flu shot vaccinations and breast exams, a result consistent with findings reported by Kenkel (1994). However, people with self-reported good health tended to reduce their use of flu shots, perhaps due to their strong natural immunity to flu. Nonetheless, self-reported healthy women were more likely to use a pap smear test, a possible indication of increasing health awareness and positive attitudes toward preventive health care. Poverty status influenced women's use of mammograms in a way that people with better financial status tended to pay more attention to regularly clinical breast care.

CONCLUSION AND FUTURE RESEARCH

This study aims at replicating and extending Xiao and Savage's (2008) research to understand the multidimensional aspect of HMOs distinguished by HMOs' consumer-friendliness, and their relationship to consumers' utilization of preventive care services. Specifically, we use ANOVA analysis to investigate the variance between Medicaid HMOs and private HMOs, and develop a dynamic equation model to capture the interrelationship between HMOs' consumer friendly characteristics and consumers' preventive care utilization. In general, our data analysis presents a similar relationship pattern as revealed by Xiao and Savage (2008): Multiform HMOs differ in their consumer-friendliness, and HMOs' consumer-friendliness imposes differential impacts on preventive care utilization. Additionally, our findings also bring important extensions to previous research in that we incorporate a longitudinal model and reveal the time-series changes of the influence of HMOs' consumer-friendliness that either the effects of early experienced HMOs' consumer-friendliness rapidly dissipate or HMOs' consumer-friendly characteristics on the concurrent terms contain most of the explanatory power.

The longitudinal approach reveals the time-series variations in the effects of HMOs' consumer-friendliness on preventive care utilization that care utilization behaviors are more likely influenced by the most recent experience than the early ones. This finding entails some practical implications to HMOs' managers that corrective actions could be taken to revise the structural design of the insurance plan, improve consumers experience, and reshape their perception and evaluation, since the care utilization behaviors are more likely influenced by the most recent experience.

Consistent with Xiao and Savage's (2008) findings, our empirical results show that HMOs' consumer-friendliness has a differential impact on preventive care variables. This state of affairs could be due to the simplified measurement of HMO characteristics by the single-item proxy indicator of the overall rating of consumers' experience with multiform HMOs. Relatively short time gap and limited number of time lags could also be one of the reasons for the mixed results, as repetitive use of preventive care rarely comes up during a short period of time, and a visibly linear distribution may gradually shape up as a function of time duration. Xiao and Savage (2008) provided additional interpretation that specialist care rather than preventive care may serve as the better dependent variable that can be significantly explained by the consumer-friendly variables, since the former demands plan approval, and more frequently relates to the cost, quality, and access boundaries of HMOs. We also note the relatively small adjusted [R.sup.2] in our proposed equation model. The possible interpretations include the following. First, health care encounters involve an interaction between patients and providers. Therefore, in order to comprehensively understand the relationship between HMO characteristics and preventive care utilization, it is necessary to consider HMO characteristics that relate to physicians and other health care providers besides consumer-oriented characteristics. Thus the omission of HMOs' structural characteristics that are provider-related could contribute to the small adjusted [R.sup.2]. Second, Miller and Luft (1994) pointed out that the performance of managed care organizations differs considerably depending on which local market areas are used for analysis. That is, the characteristics of the markets in which managed care organizations operate influence their performance significantly, and the assessment of HMO performance demands the consideration of possible contextual or contingency factors. In short, the structural form of an HMO may account for only a portion of the variance in outcomes of health care utilization; other situational factors are involved as well.

These arguments suggest the direction for future research. First, future conceptual and empirical work on HMOs should extend the notion that HMOs' structural characteristics, whether consumer- or provider-oriented, are important drivers of desired health care service utilization. For example, researchers should explore appropriate and precise measures of HMO characteristics, and consider what characteristics are associated with good performance, and what pose the potential threats to undermine the quality of care. This study presents a first step in this direction by using consumers' self-reported experience with health plans to measure HMOs' structural characteristics. Further exploration of other measurement approaches is well worth the effort. Second, we recognize the potential limitation of the study for solely depending on patients' self-reported record due to the data access limit. We drew upon MEPS data and constituted proxy indicators for our focus constructs such as HMOs' consumer-friendliness and respondents' use of preventive care. A worthwhile effort would be to complement the self-reported HMOs experience data with the actual medical records to investigate the relationship between HMOs' consumer-friendliness and actual use of preventive care. Additionally, the efforts of improving measurement could be directed to compare the self-reported data of care use with the actual medical records of care use using a Pearson-product correlation with a two-tailed test. A significant correlation coefficient would indicate the convergent validity across the two measures of preventive care utilization. Meanwhile, we also recognize the challenges of adopting actual records due to the various administrative restraints on the release of medical information. Thus, our current approach of employing MEPS data is probably the most practical and cost efficient option given the existing restraints. Third, besides HMOs' consumer-friendliness, incorporating market factors and healthcare providers-related HMOs' characteristics in the empirical analysis may present new insights into HMOs' effectiveness in terms of promoting healthcare service utilization. The differences in the effectiveness of various forms of HMOs raise questions about the mechanisms through which HMOs may operate. One possibility is that different forms of HMOs operate through different mechanisms due to the differential combination of various market and structural characteristics; and even similar outcomes may arise from very different processes due to such random interaction of various factors. Thus separate models or theories for individual forms of HMOs appear more feasible than the integrative framework of treating HMOs as a unitary plan form. It is possible that in the relationship model various market and structural characteristics may play different roles such as antecedents, consequences, mediating processes, and contextual contingencies. Therefore, it is meaningful to compare and contrast the separate emerging models of HMOs, perhaps in terms of their general dimensions of form, or some common mechanisms or processes. In this way, some convergence of theory and research on HMOs may be achieved.

Our study is more exploratory than explanatory in nature. As an exploratory study, this research provides reasonable support to advocate that research should move beyond the simple dichotomy between HMO and non-HMO insurance to differentiate the effects of specific types of insurance as well as the effects of specific care management tools, given the complexity of insurance products. To understand differences in performance among HMOs, it is important for health services researchers to reach a consensus about the HMO and benefit plan characteristics that should be routinely collected, analyzed, and discussed in reports. These are the logical next steps in understanding more about the mechanisms through which HMOs have their effects on population health. APPENDIX A Preventive Care Variables Means S.D. Variables Descriptions Year Year Year Year 2006 2007 2006 2007 BPCHEK How long since last blood pressure test 5.87 5.85 0.55 0.59 Age > 17; both genders CHOLCK How long since last blood cholesterol check 5.37 5.30 1.38 1.45 Age > 17; both genders How long since last CHECK routine check-up for 5.57 5.51 1.06 1.14 assessing overall health Age > 17; both genders FLUSHT How long since last flu 3.50 3.33 2.32 2.30 shot Age > 17; both genders PAPSMR How long since last pap smear test 5.22 5.23 1.31 1.30 Age > 17; females only BRSTEX How long since last breast exam 5.43 5.41 1.17 1.22 Age > 17; females only MAMOGR How long since last mammogram 4.50 4.41 2.06 2.07 Age > 29; females only 1 never 2 more than 5 years 3 within past 5 years 4 within past 3 years 5 within past 2 years 6 within past year Note: Preventive care utilization variables are recoded such that bigger values indicate more frequent use of preventive care.

REFERENCES

Ajzen, I., & Fishbein, M. (1980). Understanding Attitudes and Predicting Social Behavior. Englewood Cliffs, NJ: Prentice Hall.

Blaug, M. (1992). The Methodology of Economics: Or How Economics Explain. New York: Cambridge University Press.

Deaths and percentages of total deaths for the 10 leading causes of death, by race: United States, 2002. (2005). National Vital Statistics Reports. 53(17), March 7.

Duan, W., Gu, B., & Whinston, A.B. (2008). The dynamics of online word-of-mouth and product sales--an empirical investigation o the movie industry. Journal of Retailing, 84(2): 233-242.

Ellis, R. (1989). Employee choice of health insurance. Rev Econ Stat, 71(2): 215-223.

Glied, S. (2000). Managed care. In Handbook of Health Economics, vol. 1A, Newhouse J, Cuyler A (eds). North- Holland: Amsterdam; 707-753.

Herzlinger, R.E. (2004). Consumer-Driven Health Care: Implications for Providers, Payers, and Policymakers. Cambridge: Harvard Business School.

Hubbard, R., Vetter, D.E., & Little, E.L. (1998). Replication in strategic management: scientific testing for validity, generalization, and usefulness. Strategic Management Journal, 19: 243-254.

Liu, Y. (2006). Word of mouth for movies: its dynamics and impact on box office revenue. Journal of Marketing, 70: 74-89.

Kenkel, D. (1991). Health behavior, health knowledge, and schooling. J Polit Econ, 99(2): 287-305.

Kenkel, D. (1994). The demand for preventive medical care. Appl Econ, 26:313-325.

Kuhn, T. (1962). The Structure of Scientific Revolutions. Chicago: University of Chicago Press.

Lindsay, R.M. (1994). Publication system biases associated with the statistical testing paradigm. Contemporary Accounting Research, 11: 33-57.

MEPS HC-113: 2007 full year consolidated data file (November, 2009). Center for financing, access and cost trends, agency for healthcare research and quality. Rockville, MD.

MEPS HC-111: 2007 person round plan public use file (October, 2009). Center for financing, access and cost trends, agency for healthcare research and quality. Rockville, MD.

MEPS HC-105: 2006 full year consolidated data file (November, 2008). Center for financing, access and cost trends, agency for healthcare research and quality. Rockville, MD.

MEPS HC-103: 2006 person round plan public use file (November, 2008). Center for financing, access and cost trends, agency for healthcare research and quality. Rockville, MD.

Miller, R., & Luft, H. (1994). Managed care plan performance since 1980: a literature analysis. Journal of the American Medical Association. 271(19): 1512-1519.

Miller, R., & Luft, H. (1997). Does managed care lead to better or worse quality of care? Health Affairs, 16: 7-25.

Miller, R., & Luft, H. (2002). HMO performance plan update: an analysis of the literature 1997-2001. Health Affairs, 21(4): 63-86.

Popper, K. (1978). Conjectures and Refutations. London: Routledge & Kegan Paul.

President's Advisory Commission on Consumer Protection and Quality in the Health Care Industry (1998). The State of Health Care Quality: How Good is Care? Agency for Healthcare Research and Quality: Rockville, MD.

Rizzo, J.A. (2005). Are HMOs bad for health maintenance? Health Economics. Vol.14 (11).

Robinson, R., & Steiner, A. (1998). Managed Health Care. Open University Press: Buckingham.

Schauffler, H., & Rodriguez, T. (1993). Managed care for preventive services: a review of policy options. Med Care Rev, 50(2): 153-198.

Xiao, Q., & Savage, G. T. (2008). HMOs' consumer-friendliness and preventive health care utilization: exploratory findings from the 2002 MEPS. Journal of Health and Human Services Administration. Vol.31(2).

QIAN XIAO *

Eastern Kentucky University

GRANT T. SAVAGE

University of Alabama at Birmingham

WEILING ZHUANG

Eastern Kentucky University

* This research work is funded by the Killgore Research Center of the West Texas A&M University. Table 1 Demographic and Health Characteristics of Sample Respondents in Medicaid HMOs and Private HMOs Frequency and Percent Distribution of Selected Characteristic Demographic and Health Medicaid Private Difference Characteristics HMO HMO p-value (n=140) (n=479) Age 0.99 18-44 Years Old 64 (45.7) 152 (31.7) 45-64 Years Old 47 (33.6) 275 (57.4) 65-85 Years Old 29 (20.7) 52 (10.9) Race 0.59 White 92 (65.7) 355 (74.1) Black 39 (27.9) 84 (17.5) American Indian 2 (1.4) 1 (0.2) Asian 4 (2.9) 31 (6.5) Pacific Islander 0 (0.0) 3 (0.6) Multiple Races reported 3 (2.1) 5 (1.0) Education 0.00 High school and less 138 (98.6) 303 (63.3) Bachelor's degree 1 (0.7) 112 (23.4) Master's degree 0 (0.0) 50 (10.4) Doctorate degree 1 (0.7) 14 (2.9) Poverty status 0.00 Poor / negative 83 (59.3) 17 (3.5) Near poor 19 (13.6) 15 (3.1) Low income 27 (19.3) 48 (10.0) Middle income 10 (7.1) 153 (31.9) High income 1 (0.7) 246 (51.4) Perceived health status 0.00 Excellent 13 (9.3) 96 (20.0) Very good 23 (16.4) 176 (36.7) Good 44 (31.4) 160 (33.4) Fair 48 (34.3) 35 (7.3) Poor 12 (8.6) 12 (2.5) TABLE 2 Correlation Matrix of Key Variables Variables 1 2 3 4 5 1 log(Race) 1 2 log(Poverty) -.054 1 3 log(Health) -.005 .382 ** 1 4 log(Age) -.044 .040 -.185 ** 1 5 log(Education) .044 .369 ** .248 ** -.072 1 6 log(HMOs' .000 171 ** -.016 .036 -.078 consumer- friendliness) 7 log(CHOLCK) .037 .021 -.068 .262 ** -.026 8 log(CHECK) .031 .011 -.052 .102 * .021 9 log(FLUSHT) .040 .056 -.135 ** .276 ** .090 * 10 log(BPCHEK) .031 -.013 -.056 .047 .047 11 log(PAPSMR) .039 .073 145 ** -.216 ** .111 ** 12 log(BRSTEX) .018 .060 .082 * -.016 .125 ** 13 log(MAMOGR) -.009 .121 ** -.062 494 ** -.008 Variables 6 7 8 9 10 1 log(Race) 2 log(Poverty) 3 log(Health) 4 log(Age) 5 log(Education) 6 log(HMOs' 1 consumer- friendliness) 7 log(CHOLCK) .093 * 1 8 log(CHECK) .081 * .428 1 9 log(FLUSHT) -.002 .111 .167 ** 1 10 log(BPCHEK) .075 .342 .450 ** .099 * 1 11 log(PAPSMR) .071 147 .185 ** -.019 .269 ** 12 log(BRSTEX) .105 * 154 .242 ** .113 ** .296 ** 13 log(MAMOGR) .101 * .300 .165 ** .182 ** 145 ** Variables 11 12 13 1 log(Race) 2 log(Poverty) 3 log(Health) 4 log(Age) 5 log(Education) 6 log(HMOs' consumer- friendliness) 7 log(CHOLCK) 8 log(CHECK) 9 log(FLUSHT) 10 log(BPCHEK) 11 log(PAPSMR) 1 12 log(BRSTEX) 599 ** 1 13 log(MAMOGR) .178 ** .337 ** 1 Table 3 ANOVA Results of Consumer-Friendly Characteristics between Medicaid HMOs and Private HMOs Items F-value Sig. Problems in getting a doctor or nurse 9.735 0.002 Need approval for treatment 154.107 0.000 Delays in waiting for plan approval for 1.503 0.021 care Amount of information on how plan 88.529 0.000 works Problems finding information 127.663 0.000 Problems getting help from customer 0.181 0.000 service Amount of paperwork for plan 37.013 0.670 Note: 1. Significance level = .05. Table 4 ANOVA Results of Overall Rating of HMOs' Consumer- Friendliness and Related Preventive Care Utilization between Medicaid HMOs and Private HMOs Mean Variables Medicaid Private p-value HMO HMO HMOs' consumer friendliness 8.10 8.79 .00 BPCHEK 5.88 5.85 .58 CHOLCK 4.92 5.35 .03 CHECK 5.20 5.59 .05 FLUSHT 3.25 3.41 .10 PAPSMR 4.82 5.35 .00 BRSTEX 5.07 5.51 .00 MAMOGR 3.79 4.59 .00 Table 5 The Effects of HMOs' Consumer-friendliness on Preventive Care Utilization Dependent Variables Log Log Log (CHOLCK). (CHECK) (CHECK) sub.i] [log(Age).sub.i] .253 ** .090 * .258 ** [log(Race).sub.i] .053 .040 .052 [log(Education).sub.I] -.013 .034 .120 ** [log(Poverty).sub.i] .041 .037 .061 [log(Health).sub.i] -.033 -.057 .141 ** [log(HMOs' .101 * .085 * .034 consumer-friendliness. sub.it] [log(HMOs' consumer-friendliness). .021 .075 (+) -.008 sub.i.t-i] Adjusted [R.sup.2] .085 .078 .097 Log Log Log (BPCHEK) (PAPSMR) (BRSTEX) [log(Age).sub.i] .039 .193 ** .002 [log(Race).sub.i] .046 .030 .019 [log(Education).sub.I] .069 .064 .112 * [log(Poverty).sub.i] -.003 .017 -.001 [log(Health).sub.i] -.068 .086 * .054 [log(HMOs' .151 ** .007 .110 * consumer-friendliness. sub.it] [log(HMOs' consumer-friendliness). -.069 -.058 .082 * sub.i.t-i] Adjusted [R.sup.2] .017 .057 .093 Log (MAMOGR) [log(Age).sub.i] .487 ** [log(Race).sub.i] .025 [log(Education).sub.I] -.013 [log(Poverty).sub.i] .113 ** [log(Health).sub.i] -.013 [log(HMOs' .104 * consumer-friendliness. sub.it] [log(HMOs' consumer-friendliness). .080 (+) sub.i.t-i] Adjusted [R.sup.2] .149 Note: (1.) (+) p < .1; * p < .05; ** p < .01. (2.) Outliers were deleted when they are out of 3 standard deviations.