Zero-truncated Negative Binomial Research Articles

Analysis of Railroad Accident Prediction using Zero-truncated Negative Binomial Regression and Artificial Neural Network Model: A Case Study of National Railroad in South Korea

Conventional statistical models, such as Poisson or negative binomial, have predefined underlying relationships between explanatory variables. However, artificial neural network (ANN), which overcomes the limitations of statistical prediction model, have gained popularity in practice and research for their ability to increase prediction accuracy. Thus, this study employs zero-truncated negative binomial (ZTNB) models and artificial neural network (ANN) models to analyze the distribution of railroad accident frequency and the corresponding number of casualties for 1995–2021 accident dataset of Korea's national railroad. The study mainly focused on two most dominant accident types which are human-involved accidents (accounted for 89.2% of all accidents) and ground-level crossing accidents (9.6%) from the historic dataset. This is because not just data proportion, rather such accident types were received very little attention compared to fatal train accident in the accident prediction study. Further, these types of accident showed clearly tended to decrease over time, but time trend has been found very weak at the type of fatal train accident. The performance of the developed models was estimated by mean square error (MSE), the root mean square error (RMSE), and the coefficient of determination (R2). Results present that ANN models outperform ZTNB models in fitting and prediction, demonstrating once again ANN's superiority over statistical models for predicting accident frequency and casualty count.

Understanding online review behaviors of patients in online health communities: an expectation-disconfirmation perspective

PurposeAlthough online health communities (OHCs) and online patient reviews can help to eliminate health information asymmetry and improve patients' health management, how patients write online reviews within OHCs is poorly understood. Thus, it is very necessary to determine the factors influencing patients' online review behavior in OHCs, including the emotional response and reviewing effort.Design/methodology/approachBased on expectation-disconfirmation theory, this study proposes a theoretical model to analyze the effects of service quality perception (i.e. outcome quality and process quality perceptions) and disconfirmation (i.e. outcome quality and process quality disconfirmations) on patients' emotional response and reviewing effort. The authors test the research model by using empirical data collected from a popular Chinese OHC and applying ordinary least squares (OLS) regression and zero-truncated negative binomial (ZTNB) regression models.FindingsBoth service quality perception and disconfirmation have a positive effect on patients' positive emotional intensity in textual reviews, and disease severity enhances these relationships of process quality. Moreover, there is an asymmetric U-shaped relationship among service quality perception, disconfirmation and reviewing effort. Patients who perceive low service quality have higher reviewing effort, while service quality disconfirmation has the opposite relationship. Specifically, patients' effort in writing textual reviews is lowest when perceived outcome quality is 3.5 (on a five-point scale), perceived process quality is 4 or outcome quality and process quality disconfirmations are −1.Originality/valueThis study is the first to examine patients' online review behavior and its motivations and contributes to the literature on online reviews and service quality. In addition, the findings of this study have important management implications for service providers and OHC managers.

Power of Overdispersion Tests in Zero-Truncated Negative Binomial Regression Model

Poisson regression is the most extensively used model for modeling data that are measured as counts. The main characteristic of Poisson regression model is the equidispersion limitation in which the mean and variance of the count variable are the same. However, in many situations the variance of the count variable is greater than the mean which causes overdispersion, and hence, poor fit will be resulted when inference about regression parameters. Alternatively, the negative binomial regression is preferred when overdispersion is present. In addition, in particular cases, the zero counts are not observed in data which is known as zero-truncation. In the presence of overdispersion in zero-truncated count data, the zero-truncated negative binomial (ZTNB) regression model can be used as an alternative to zero-truncated Poisson (ZTP) regression model. In this paper, for testing overdispersion in ZTNB regression model against ZTP regression model, the likelihood ratio test (LRT), score test, and Wald test are proposed. A Monte-Carlo simulation is carried out in order to examine the empirical power for statistics of these tests under different levels of overdispersion and various sample sizes. The simulation results indicate that Wald test is more powerful than the LRT and score test for detecting the overdispersion parameter in ZTNB regression model against ZTP regression model, since it provides the highest statistical power. Thus, the Wald test is preferable for detecting the overdispersion problem in zero-truncated count data.

Emotions Behind Drive-by Download Propagation on Twitter

Twitter has emerged as one of the most popular platforms to get updates on entertainment and current events. However, due to its 280-character restriction and automatic shortening of URLs, it is continuously targeted by cybercriminals to carry out drive-by download attacks, where a user’s system is infected by merely visiting a Web page. Popular events that attract a large number of users are used by cybercriminals to infect and propagate malware by using popular hashtags and creating misleading tweets to lure users to malicious Web pages. A drive-by download attack is carried out by obfuscating a malicious URL in an enticing tweet and used as clickbait to lure users to a malicious Web page. In this article, we answer the following two questions: Why are certain malicious tweets retweeted more than others? Do emotions reflecting in a tweet drive virality? We gathered tweets from seven different sporting events over 3 years and identified those tweets that were used to carry to out a drive-by download attack. From the malicious ( N = 105, 642) and benign ( N = 169, 178) data sample identified, we built models to predict information flow size and survival. We define size as the number of retweets of an original tweet, and survival as the duration of the original tweet’s presence in the study window. We selected the zero-truncated negative binomial (ZTNB) regression method for our analysis based on the distribution exhibited by our dependent size measure and the comparison of results with other predictive models. We used the Cox regression technique to model the survival of information flows as it estimates proportional hazard rates for independent measures. Our results show that both social and content factors are statistically significant for the size and survival of information flows for both malicious and benign tweets. In the benign data sample, positive emotions and positive sentiment reflected in the tweet significantly predict size and survival. In contrast, for the malicious data sample, negative emotions, especially fear, are associated with both size and survival of information flows.

Introducing Hurdle_T as a Count Regression Model with High Potency in Solving Problems

Count data regression models exhibit different strengths and weaknesses in their bids to solving problems. The study considers six count models namely Poisson Regression Model (PRM), Negative Regression Model (NBRM), Zero Inflated Poisson (ZIP), Zero Inflated Negative Binomial (ZINB), Zero Truncated Poisson (ZTP) and Zero Truncated Negative Binomial (ZTNB) and an additional model called hurdle_T. These models are used to analyze two health data sets. The data on male breast cancer reveals that male breast cancer cuts across all age brackets or categories but it is more prevalent between the ages of 50 and 60. The PRM yields a better result than the NBRM in the case of cancer data as shown by the information criteria. The analysis of the second data, which is on doctor’s visit reveals that ZINB yields a better result than the other five models, followed by NBRM, then the ZTNB before their Poisson counterparts. The hurdle_T model shows the propensity of each coefficient as reflected by the positive count in the Tobit (Binary) model. The study also shows that at 65 years and above, gender has significant effect on doctor’s visit. In particular, females, more than males attract more doctors’ visit in the said age range. Government policies should provide more funds in the health sector to accommodate cancer cases in terms of the provision of awareness, studies/ research and infrastructural development. Males should be encouraged to visit clinics especially in their late forties and above for breast cancer related checkup. At age 65 and above, doctors visit to patients are frequent, especially to females. Policy of government in the health sector should accommodate a favourable adjustment in the budget to take care of doctors’ visit.

Modelling Deaths Associated with Road Traffic Accidents and other Factors on Great North Road in Zambia between the Years 2010 and 2016 Using Poisson Models

Background: According to the World Health Organization (WHO), 1.24 million people die annually on the world’s roads, with 20-50 million sustaining non-fatal injuries. More than 85% (1.05 million) of the global deaths due to injuries occur in the developing world. Road traffic deaths and injuries are a major but neglected public health challenge that requires concerted efforts for effective and sustainable prevention. The objectives of the study were to estimate the incidence rate of death from RTAs, to determine factors associated with serious and fatal Road Traffic Accidents (RTAs) and to determine which of the poisson models fit the count data better. Methods: Data was collected from Zambia Police (ZP), Traffic Division on accidents that occurred on the Great North Road (GNR) highway between Lusaka and Kapiri-Mposhi in Zambia from January 1, 2010 to December 31, 2016. Results from standard Poisson regression were compared to those obtained using the Negative Binomial (NB), Zero-Truncated Negative Binomial (ZTNB) and the Zero-Truncated Poisson (ZTP) regression models. Diagnostic tests were used to determine the best fit model. The data was analysed using STATA software, version 14.0 SE (Stata Corporation, College Station, TX, USA). Results: A total of 1, 023 RTAs were analysed in which 1, 212 people died. Of these deaths, 82 (7%) were Juveniles and 1, 130 (93%) were adults. Cause of accident such as pedestrians crossing the road accounted for 30% (310/1,023) while 29% (295/1,023) were as a result of driver’s excessive speed. The study revealed that driving in the early hours of the day (1AM-6AM) as compared to driving in the night (7PM-12AM) had a significant increase in the incidence rate of death from RTAs, Incidence Rate Ratio (IRR) of 2.1, (95% CI={1.01-4.41}), p-value=0.048. Results further showed that public transport as compared to private transport had an increased incidence rate of death from RTAs (IRR=5.65, 95% CI={2.97-10.73}), p-value<0.0001. The two competing models were the ZTP and the ZTNB. The ZTP had AIC=1304.55, BIC= 1336.55, whereas the ZTNB had AIC=742.25 and BIC=819.69. This indicated that the ZTNB with smaller AIC and BIC was the best fit model for the data. Conclusion: There is a reduced incidence of dying if one is using a private vehicle as compared to a public vehicle. Driving in the early hours of the day (1AM and 6AM) had an increased incidence of death from RTAs. This study suggests that when dealing with counts in which there are a few zeros observed such as in serious and fatal RTAs, ZTNB fits the data well as compared to other models.

Road Infrastructure Impacts on Transportation Safety (Case Study: Yazd Rural Roads)

Background: Statistics showed that road traffic fatalities is amongst the top ten cause of. Road infrastructure quality play a vital role in reducing traffic related accidents. There are different ways to inspect road infrastructures. In Iran however road safety inspections are mostly based on periodic field surveys. This study goaled to investigate the importance of road infrastructure quality on transportation yearly crash rate index based on 3 years (2015-2017) police reported crash data with the case study of Yazd province. Infrastructure quality data are collected from Yazd department of transportation and road maintenance for Yazd rural highways. The recorded qualities were based on 4 point scale road deficit (i.e. very low, low, medium and high). The influence of road infrastructure quality were examined based on both descriptive statistics and crash rate index modeling. It was tried to compare ten different modeling structure based on akaike information criterion (AIC) and Bayesian information criterion (BIC). The models encompasses: Poisson, negative binomial (NB2), zero-truncated-poisson (ZTP), zero-truncated-negative-binomial (ZTNB), zero-inflated-poisson (ZIP), zero-inflated-negative-binomial (ZINB), REPoisson, RENB2, FEPoisson and FENB. Regardless of the yearly recorded road infrastructure data, investigations demonstrated that nearly 75% of road infrastructure qualities are in medium condition for all years. Results: Results showed that the FENB model outperform the competing models. Results also showed the reduction in road traffic crash rate index when road infrastructure quality decreased by one level form very low to low and also from low to medium. But there seems that the index will increase when road infrastructure quality decreased one level from medium to high. Conclusion: The whole results claims that we are in critical condition for our road infrastructure maintenance budget assignment.

Recreational value and optimal pricing of national parks: lessons from Maasai Mara in Kenya

ABSTRACTThis paper estimates the recreational value and optimal pricing for recreation services in the Maasai Mara National Park in Kenya. To achieve this objective, data from 323 Park visitors were collected. Single-site individual travel cost method (ITCM) using count data models [zero truncated Poisson (ZTP), zero truncated negative binomial (ZTNB), negative binomial with endogenous stratification (NBSTRAT), and Poisson with endogenous stratification (PSTRAT)] was applied. Results show a consumer surplus of US$ 115 per visitor per day, which translates to a Park recreational value of is US$ 73.076 million per year. The optimal conservation fee that would maximize revenue for the Park was estimated at US$ 86.90 per day, a value which is less than the consumer surplus. To maximize revenue, the Park managers can therefore hike the price to capture some consumer surplus or invest in substitute facilities to increase expenditure at the site (onsite costs). It's also important to note that the value estimated in this study is for recreation only. There are other ecosystem services (provisioning, regulating, cultural and supporting) which are produced in the Park. These should also be captured and paid for, so as to avail more funds for conservation and production of more ecosystem services.

Exploring the effects of state highway safety laws and sociocultural characteristics on fatal crashes

ABSTRACTObjective: Distinguished from the traditional perspectives in crash analyses, which examined the effects of geometric design features, traffic factors, and other relevant attributes on the crash frequencies of roadway entities, our study focuses on exploring the effects of highway safety laws, as well as sociocultural characteristics, on fatal crashes across states.Methods: Law and regulation related data were collected from the Insurance Institute for Highway Safety, State Highway Safety Offices, and Governors Highway Safety Association. A variety of sociodemographic characteristics were obtained from the U.S. Census Bureau. In addition, cultural factors and other attributes from a variety of resources are considered and incorporated in the modeling process. These data and fatal crash counts were collected for the 50 U.S. states and the District of Columbia and were analyzed using zero-truncated negative binomial (ZTNB) regression models.Results: The results show that, in law and regulation–related factors, the use of speed cameras, no handheld cell phone ban, limited handheld cell phone ban, and no text messaging ban are found to have significant effects on fatal crashes. Regarding sociocultural characteristics, married couples with both husband and wife in the labor force are found to be associated with lower crash frequencies, the ratios of workers traveling to work by carpool, those driving alone, workers working outside the county of residence, language other than English and limited English fluency, and the number of licensed drivers are found to be associated with higher crash frequencies.Conclusions: Through reviewing and modeling existing state highway safety laws and sociocultural characteristics, the results reveal new insights that could influence policy making. In addition, the results would benefit amending existing laws and regulations and provide testimony about highway safety issues before lawmakers consider new legislation.

One-inflation and unobserved heterogeneity in population size estimation.

We present the one-inflated zero-truncated negative binomial (OIZTNB) model, and propose its use as the truncated count distribution in Horvitz-Thompson estimation of an unknown population size. In the presence of unobserved heterogeneity, the zero-truncated negative binomial (ZTNB) model is a natural choice over the positive Poisson (PP) model; however, when one-inflation is present the ZTNB model either suffers from a boundary problem, or provides extremely biased population size estimates. Monte Carlo evidence suggests that in the presence of one-inflation, the Horvitz-Thompson estimator under the ZTNB model can converge in probability to infinity. The OIZTNB model gives markedly different population size estimates compared to some existing truncated count distributions, when applied to several capture-recapture data that exhibit both one-inflation and unobserved heterogeneity.

International tourists’ length of overnight stay in Venice

This study concerns factors influencing international tourists’ length of overnight stay in Venice. The data are from a survey conducted by the Bank of Italy on international tourism. Both zero-truncated negative binomial (ZTNB) and ordinary least squares (OLS) regressions are estimated. In general, the estimates from ZTNB and OLS regressions are similar. The results suggest that age, returning directly to the country of residence and the summer season have a positive influence on the length of stay. On the other hand, international tourist expenditure and visiting other places than Venice have a negative impact on the length of stay. Moreover, the average length of overnight stays differs among nationalities and those who have visited Venice previously tend, on average, to stay longer. One finding is that visitors from North America and Australia stay longer and spend more than visitors coming from neighbouring countries.

Abstract P3-01-05: The use of Tc-99 tilmanocept in sentinel lymph node biopsy after neoadjuvant chemotherapy in clinically node-negative patients with breast cancer

Abstract Background: Neoadjuvant chemotherapy (NAT) has been shown to induce fibrosis and inflammation that alters lymphatic drainage of axillary lymph nodes in breast cancer. Technetium- 99 Tilmanocept (TcTM), a CD206-macrophage receptor targeted radiopharmaceutical, is a small agent with recent FDA-approval for lymphatic mapping. No prior studies have investigated the use of TcTM in the neoadjuvant setting. The aim of this study was to compare the identification rate, node-positivity rate, and number of total nodes evaluated in sentinel lymph node (SLN) biopsy with TcTM and vital blue dye (VBD) in clinically node-negative patients receiving NAT vs. initial surgery. Methods: A retrospective review was conducted on patients undergoing SLN biopsy with TcTM and VBD from May 2013- May 2015 at UCSD. Patients with a history of prior SLN biopsy or axillary lymph node dissection were excluded. Patients undergoing neoadjuvant chemotherapy or receiving > 3 months of neoadjuvant endocrine therapy were grouped and compared to patients undergoing initial surgical treatment. The SLN identification and node-positivity rates were compared with the X2 test. To compare the number of SLNs evaluated between groups, a zero-truncated negative binomial (ZTNB) count model was constructed to assess the effect of NAT and other covariates on the SLN count. Covariates included age, body mass index (BMI), gender, surgeon, mastectomy vs. lumpectomy, node positivity, pathologist, T-stage, and receptor status. A p-value < 0.05 was used for statistical significance. Results: Of the 417 total SLN cases identified, 72 (17.2%) cases were in patients who had received NAT (61- chemo, 11- endocrine). The SLN identification rate was 100% in both groups (p= 1.0). Overall, there were 68 (16.3%) cases of SLN-positivity, 14 (19.4%) in the NAT group versus 54 (15.7%) in the non-NAT group (p= 0.54). The median number of identified nodes was 3 in both groups. In the ZTNB count model, age, surgeon and evaluating pathologist were significant predictors of the total number of SLN evaluated. The use of NAT did not significantly affect the number SLNs evaluated. Incident rate ratios, confidence intervals and p-values are reported in the attached table. Sentinel Lymph Node Count ModelVariabeIRR95% CI LL95% CI ULp-valueAge per 5 years0.960.930.990.03Surgeon #21.231.051.450.01NAT1.140.921.410.22Pathologist #20.720.570.900.005Pathologist #31.010.841.220.90Pathologist #40.930.661.320.70IRR: incident rate ratio, NAT: neoadjuvant chemoendocrine therapy, CI: confidence interval, LL: lower limit, UL: upper limit Discussion: Prior studies have indicated that NAT may induce fibrosis and inflammation that may obscure lymphatic mapping procedures. For SLN biopsy with TcTM in VBD in our study, the use of NAT did not change the identification rate or node-positivity rate. Additionally, when controlling for covariates, the use of NAT did not change the total number of SLNs evaluated. While NAT might induce fibrosis and inflammation, SLN biopsy with TcTM and VBD is technically successful in clinically node-negative patients undergoing neoadjuvant chemotherapy. Citation Format: Unkart JT, Wallace AM. The use of Tc-99 tilmanocept in sentinel lymph node biopsy after neoadjuvant chemotherapy in clinically node-negative patients with breast cancer. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P3-01-05.

A Cross-Model Comparison of Travel Time Inclusion Techniques in Recreational Fishing Demand Analysis

Incorporating travel time into recreational demand analysis is a much debated topic. The aim of this paper is to contribute to discourse surrounding the travel cost method by analysing data from a bass angling tournament, the Amatola Bass Classic, held annually in the Eastern Cape Province of South Africa. Firstly, data collected from 64 respondents were used to test the goodness-of-fit of four models: the Standard Poisson, the Zero-Truncated Poisson, the Negative Binomial and the Zero-Truncated Negative Binomial. These models were then utilised to gauge the difference between two standard ways of treating time, by including it both as a separate parameter and as an intrinsic part of travel costs. The data were found to be overdispersed, pointing to a greater level of reliability in the Negative Binomial Models. Using all four models it was shown that the incorporation of time into the overall cost of the trip produced results which were better able to explain the amount of trips taken by anglers than those model specifications where time was included as a separate parameter.

Tweeting the terror: modelling the social media reaction to the Woolwich terrorist attack

Little is currently known about the factors that promote the propagation of information in online social networks following terrorist events. In this paper we took the case of the terrorist event in Woolwich, London in 2013 and built models to predict information flow size and survival using data derived from the popular social networking site Twitter. We define information flows as the propagation over time of information posted to Twitter via the action of retweeting. Following a comparison with different predictive methods, and due to the distribution exhibited by our dependent size measure, we used the zero-truncated negative binomial (ZTNB) regression method. To model survival, the Cox regression technique was used because it estimates proportional hazard rates for independent measures. Following a principal component analysis to reduce the dimensionality of the data, social, temporal and content factors of the tweet were used as predictors in both models. Given the likely emotive reaction caused by the event, we emphasize the influence of emotive content on propagation in the discussion section. From a sample of Twitter data collected following the event (N = 427,330) we report novel findings that identify that the sentiment expressed in the tweet is statistically significantly predictive of both size and survival of information flows of this nature. Furthermore, the number of offline press reports relating to the event published on the day the tweet was posted was a significant predictor of size, as was the tension expressed in a tweet in relation to survival. Furthermore, time lags between retweets and the co-occurrence of URLS and hashtags also emerged as significant.

Analysis of U.S. freight-train derailment severity using zero-truncated negative binomial regression and quantile regression

Derailments are the most common type of freight-train accidents in the United States. Derailments cause damage to infrastructure and rolling stock, disrupt services, and may cause casualties and harm the environment. Accordingly, derailment analysis and prevention has long been a high priority in the rail industry and government. Despite the low probability of a train derailment, the potential for severe consequences justify the need to better understand the factors influencing train derailment severity. In this paper, a zero-truncated negative binomial (ZTNB) regression model is developed to estimate the conditional mean of train derailment severity. Recognizing that the mean is not the only statistic describing data distribution, a quantile regression (QR) model is also developed to estimate derailment severity at different quantiles. The two regression models together provide a better understanding of train derailment severity distribution. Results of this work can be used to estimate train derailment severity under various operational conditions and by different accident causes. This research is intended to provide insights regarding development of cost-efficient train safety policies.

Individual and County-Level Factors Associated with Use of Multiple Prescribers and Multiple Pharmacies to Obtain Opioid Prescriptions in California

Use of multiple prescribers and pharmacies is a means by which some individuals misuse opioids. Community characteristics may be important determinants of the likelihood of this phenomenon independent of individual-level factors. This was a retrospective cohort study with individual-level data derived from California's statewide prescription drug monitoring program (PDMP) and county-level socioeconomic status (SES) data derived from the United States Census. Zero-truncated negative binomial (ZTNB) regression was used to model the association of individual factors (age, gender, drug schedule and drug dose type) and county SES factors (ethnicity, adult educational attainment, median household income, and physician availability) with the number of prescribers and the number of pharmacies that an individual used during a single year (2006). The incidence rates of new prescriber use and new pharmacy use for opioid prescriptions declined across increasing age groups. Males had a lower incidence rate of new prescriber use and new pharmacy use than females. The total number of licensed physicians and surgeons in a county was positively, linearly, and independently associated with the number of prescribers and pharmacies that individuals used for prescription opioids. In summary, younger age, female gender, and living in counties with more licensed physicians and surgeons were associated with use of more prescribers and/or more pharmacies for obtaining prescription opioids.

A Bayesian zero-truncated approach for analysing capture–recapture count data from classical scrapie surveillance in France

Capture–recapture (CR) methods are used to study populations that are monitored with imperfect observation processes. They have recently been applied to the monitoring of animal diseases to evaluate the number of infected units that remain undetected by the surveillance system. This paper proposes three Bayesian models to estimate the total number of scrapie-infected holdings in France from CR count data obtained from the French classical scrapie surveillance programme. We fitted two zero-truncated Poisson (ZTP) models (with and without holding size as a covariate) and a zero-truncated negative binomial (ZTNB) model to the 2006 national surveillance count dataset. We detected a large amount of heterogeneity in the count data, making the use of the simple ZTP model inappropriate. However, including holding size as a covariate did not bring any significant improvement over the simple ZTP model. The ZTNB model proved to be the best model, giving an estimation of 535 (CI95% 401–796) infected and detectable sheep holdings in 2006, although only 141 were effectively detected, resulting in a holding-level prevalence of 4.4‰ (CI95% 3.2–6.3) and a sensitivity of holding-level surveillance of 26% (CI95% 18–35). The main limitation of the present study was the small amount of data collected during the surveillance programme. It was therefore not possible to build complex models that would allow depicting more accurately the epidemiological and detection processes that generate the surveillance data. We discuss the perspectives of capture–recapture count models in the context of animal disease surveillance.

Hospitalization in Tajikistan: determinants of admission, length of stay, and out‐of‐pocket expenditures. Results of a national survey

To assess factors explaining hospitalization in Tajikistan and discuss policy implications for reforms in hospital care. This study involves a secondary analysis of micro-data collected from a nationally-representative household survey conducted in Tajikistan in 2003. Three empirical models are employed: binomial logit regression for the admission to the hospital; zero truncated negative binomial (ZTNB) regression for the length of hospital stay; and ordinary least square (OLS) for the amount of out-of-pocket expenditures for hospitalization. Variation in hospital admission is due to the differences in ability to pay, long-standing illness, gender, age, and educational level. Factors explaining out-of-pocket expenditures include ability to pay, having long-standing illness, and having surgery and receiving intensive care. The most important out-of-pocket expenditures are payments for pharmaceuticals and supplies. Finally, long hospital stay that is the result of outdated treatment protocols, rigid financial and management system, lack of funding, and weakness of primary and long-tem care. As a result, long time in inpatient care is mostly used ineffectively. Strategies to address the existing deficiencies include voluntary community-based health insurance for rural areas, targeted subsidized care for the neediest, improvements in procurement of pharmaceuticals and supplies, and rationalization of hospital primary and long-term care.