Mixed Logit Model Research Articles

Consumer willingness to pay (WTP) for beef based on color and price discounts

Discoloration of beef products leads retailers to offer discounts or discard beef products, as consumers tend to reject a brownish beef color. This practice contributes to retail food waste and inefficient resource allocation linked to beef production. The objective of this study was to investigate U.S. consumer preferences for beef color and price discounts. After an initial test for screen acuity regarding color perception using a condensed form the ishihara test, data on meat purchasing behavior as well as demographic information was collected for all participants. An additional part of the study undertook a hypothetical discrete choice experiment using digital photos of beef steaks with different meat color, price, and discount labels in retail packaging to examine consumer preferences, willingness to pay, and the effects of discount pricing. Results of a Mixed Logit Model based on choice data show that beef shoppers perceive, see, and prefer color nearly linearly across days of retail display. Consumers' willingness to pay was negative for 4, 7 and 9 days of retail display relative to day zero and decreased with an increase in discoloration. Results indicate that beef shoppers would have to be financially compensated for choosing a steak that was discolored after 9 days of retail display. However, male beef buyers exhibit weaker preferences for color and could be a target segment for promoting discolored beef products to reduce meat waste.

Patient and clinician preferences for biologic treatments for severe uncontrolled asthma: a discrete choice experiment (DCE)

Objectives To estimate the preferences of patients with asthma and asthma-treating clinicians for attributes of biologic treatments, to compare patients’ and clinicians’ preferences, and to better understand the reasons for their preferences. Methods Adults with moderate-to-severe asthma and clinicians who treat asthma in the US completed a cross-sectional, online survey including a discrete choice experiment (DCE) that consisted of seven attributes spanning treatment efficacy, risk and convenience. Marginal utilities were estimated using a mixed logit model, and relative attribute importance scores calculated. Clinicians were also asked about the value of biomarker agnostic biologic treatments. The survey was followed by qualitative interviews targeting a sub-sample of survey participants, in which the rationale behind their survey responses was discussed. Results In the DCE, both patients and clinicians placed the most importance on exacerbation and hospitalization rate reduction, and risk of injection site reaction. Patients valued location of administration more than clinicians. Rationale for individual-level preferences varied, with patients and clinicians reporting their preference depended on event frequency and anticipated quality of life impacts. Clinicians mentioned compliance and financial impacts, while patients mentioned personal experience, particularly around site reactions. Most patients and clinicians would value a biomarker agnostic asthma treatment. Conclusions Asthma treatment preferences are largely driven by treatment efficacy and minimizing the risk of site reactions, although preferences differ between patients and clinicians across other attributes, highlighting the need for shared decision-making and individualized care.

Scale and rate heterogeneity in the EQ-5D-5L valuation

ObjectivesTo estimate values on a quality-adjusted life year (QALY) scale using individual preference evidence, choice analyses typically include ancillary parameters, such as scale factors and discount rates. These parameters potentially differ among respondents. In this study, we investigated how allowing heterogeneity in scale and rate affects the estimation of EQ-5D-5L values.MethodsUsing the first wave of the 2016 EQ-5D-5L valuation study (N = 1017), we estimated a conditional logit (CL) model and three mixed logit models: random scale, random rate, and bivariate. Prior to the exploratory study, we hypothesized that scale and rate are correlated and that allowing heterogeneity in both parameters decreases the number of insignificant incremental effects. We confirmed the exploratory findings by re-estimating these models using paired comparison responses from a second wave (N = 1229).ResultsScale and rate exhibited significant heterogeneity and were positively correlated. As hypothesized, allowing this heterogeneity improved the face validity of the EQ-5D-5L value set by reducing the number of insignificant incremental effects (from 6 to 2 p-values > 0.05; out of 20). Nevertheless, the CL and bivariate mixed logit estimates are highly correlated and concordant (Pearson correlation coefficient of 0.897, Spearman correlation coefficient of 0.888, Lin’s concordance coefficient of 0.763).ConclusionsAllowing this heterogeneity adds three parameters to the estimation (two variances and a correlation) and improves the face validity of the EQ-5D-5L values. This finding may influence experimental design and choice analysis in health valuation more generally.

The spatial planning of housing and urban green space: A combined stated choice experiment and land-use modeling approach

Urban green space (UGS) is receiving increasing attention as a means to make cities better adapted to climate change and to create high-quality living environments for citizens. The spatial planning of UGS generally asks a trade-off between optimizing land-use and accessibility characteristics which may differ between housing types. In this paper, we use a stated choice experiment to estimate residential location preferences related to neighborhood land-use, UGS and accessibility characteristics in the framework of a land-use model. A national sample of 394 Dutch homeowners participated in the experiments and mixed logit models were estimated by housing type distinguishing detached, row houses and apartments. The estimation results show that trade-offs made between a green neighborhood and accessibility to urban amenities tend to differ between the housing types. The estimates are used as parameters in a housing land-use allocation model. An application of the model to a residential area development problem shows that creating green buffers along road infrastructure is the most beneficial way of allocating UGS considering the housing value for residents. The application further shows that apartments benefit more strongly from UGS in the neighborhood than from high accessibility of urban amenities compared to detached and row houses. Finally, we find that the optimal spatial allocation of UGS depends on whether maximizing housing market-value or residents’ utility is the prime objective.

Elicitation of societal preferences for chronic lymphocytic leukemia’s treatments: a discrete choice experiment

The overall value of treatments for chronic lymphocytic leukemia (CLL) depends on several factors, including preferences of the general population, who contributes to the financing of health systems. This study investigated societal preferences for attributes of CLL treatments in Italy. An online large-scale survey was designed using a discrete choice experiment (DCE) methodology and delivered to the Italian adult general population. Ten treatment attributes were identified, covering efficacy, safety, operational aspects and (hypothetical) out-of-pocket cost. DCE data were analyzed using a mixed logit regression model, estimating the willingness-to-pay for attribute levels’ change. The general population significantly preferred more effective treatments, with shorter duration, administered orally rather than orally + intravenously. Changes in therapy duration, frequency of checkups and organ damage risk had the greatest impact on preferences. The integration of societal preferences in the value judgments of CLL therapies may help health authorities in establishing priority setting and taking pricing-reimbursement decisions.

Eliciting psychiatric nurses' preferences for workplace violence prevention: a protocol for discrete choice experiment.

Workplace violence against healthcare workers has become a serious global public health problem. The incidence of workplace violence towards Psychiatric nurses is higher than in all other medical institutions, up to 84.2% per year. It not only negatively affects many aspects of healthcare workers' lives, but also destroys the harmony of the nurse-patient relationship and reduces the quality of nursing care. The number of psychiatric nurses in China was approximately 96,000, far lower than most other countries and unable to meet the growing demand for mental health. However, the increase in workplace violence has future exacerbates the current shortage of nurses. Therefore, it is necessary to develop effective strategies to prevent psychiatric nurses from suffering from workplace violence, thereby to reduce nurse turnover and improve the quality of nursing care. A comprehensive understanding of psychiatric nurses' preferences and priorities for preventing workplace violence is an important prerequisite before formulating strategies and taking measures. Unfortunately, to date, no research has investigated the psychiatric nurses' preferences. Therefore, a discrete choice experiment (DCE) is conducting to explore the psychiatric nurses' preferences for workplace violence prevention. This article reports on methodological details of the DCE. Six attributes were developed through a literature review, one-on-one interviews and focus group discussions. D-efficient design in NGENE was used to generate choice sets. SPSS 24.0 will be used for descriptive analysis of social Demography, and Stata 16.0 will be used for analysis of DCE data. A multinomial logit model will be used to preliminarily explore trade-offs between workplace violence prevention characteristics included in the choice tasks. Then, in a mixed logit model, we plan to choose some arbitrarily defined base violence prevention program and will use the nlcom command to evaluate the probability of an alternative violence prevention program. The study was approved by the relevant ethics committees. Our findings will emphasize priority intervention areas based on the preferences of psychiatric nurses and provide references for hospitals to develop and improve workplace violence prevention strategies. The results will be shared through seminars, policy briefs, peer-reviewed journal articles and online blogs.

The EQ-5D-5L valuation study for Trinidad and Tobago

PurposeThe 2016 EQ-5D-3L value set for Trinidad and Tobago (T&T) allows for the calculation of EQ-5D-5L values via the crosswalk algorithm. The 2016 value set was based on methods predating the EQ-VT protocol, now considered the gold standard for developing EQ-5D value sets. Furthermore, direct elicitation of EQ-5D-5L is preferred over crosswalked values. This study aimed to produce an EQ-5D-5L value set for T&T.MethodsA representative sample (age, sex, geography) of adults each completed 10 composite Time Trade-Off (cTTO) tasks and 12 Discrete Choice Experiment (DCE) tasks in face-to-face interviews. The cTTO data were analyzed using a Tobit model that corrects for heteroskedasticity. DCE data were analyzed using a mixed logit model. The cTTO and DCE data were combined in hybrid models.ResultsOne thousand and seventy-nine adults completed the valuation interviews. Among the modelling approaches that were explored, the hybrid heteroskedastic Tobit model produced all internally consistent, statistically significant coefficients, and performed best in terms of out-of-sample predictivity for single states. Compared to the existing EQ-5D-5L crosswalk set, the new value set had a higher number of negative values (236 or 7.6% versus 21 or 0.7%). The mean absolute difference was 0.157 and the correlation coefficient between the two sets was 0.879.ConclusionThis study provides a value set for the EQ-5D-5L for T&T using the EQ-VT protocol. We recommend this value set for QALY computations relating to T&T.

Patient preferences for generic substitution policies: a discrete choice experiment in China.

Background: Generic substitution policies have been widely implemented worldwide to enhance the accessibility of medications. Nevertheless, certain patients have voiced discontent with these policies. This study aimed to evaluate the patient preferences for generic substitution policies and explore the potential for optimization to enhance patient acceptance. Methods: A discrete choice experiment (DCE) was conducted to estimate the relative importance (RI) of five attributes, including generic consistency evaluation (GCE), reimbursement rate, medication use control, information disclosure, and post-marketing surveillance. Respondents were recruited among inpatients and outpatients in three cities and surveys were conducted face-to-face. Preference coefficients, RI of attributes, and the uptake rate of various policies were computed using a mixed logit model. The interaction effects were also included to examine preference heterogeneity. Results: A total of 302 patients completed the survey. All five attributes significantly impacted policy acceptance. GCE held the highest RI value at 56.64%, followed by reimbursement rate (RI = 12.62%), information disclosure (RI = 12.41%), post-marketing surveillance (RI = 9.54%), and medication use control (RI = 8.80%). Patient preferences varied depending on their gender and income. The patient uptake rate of China's current policy was only 68.56%. If all generics were to pass GCE without altering the other attributes, the uptake rate of policies would rise to 82.63%. Similarly, implementing information disclosure without changing other attributes would result in a 78.67% uptake rate, which is comparable to the effect of a 10% increase in reimbursement rate for generics (78.81%). Combining these policies could mitigate the adverse effects of mandatory substitution on patient. Conclusion: Chinese patient preferences for generic substitution policies were mainly influenced by GCE. China's current generic substitution policy has room for further optimization to enhance patient acceptance.

Estimating an EQ-5D-Y-3L Value Set for Brazil.

The EQ-5D-Y-3L is a generic measure of health-related quality of life in children and adolescents. Although the Brazilian-Portuguese EQ-5D-Y-3L version is available, there is no value set for it, hampering its use in economic evaluations. This study aimed to elicit a Brazilian EQ-5D-Y-3L value set based on preferences of the general adult population. Two independent samples of adults participated in an online discrete choice experiment (DCE) survey and a composite time trade-off (cTTO) face-to-face interview. The framing was "considering your views for a 10-year-old child". DCE data were analyzed using a mixed-logit model. The 243 DCE predicted values were mapped into the observed 28 cTTO values using linear and non-linear mapping approaches with and without intercept. Mapping approaches' performance was assessed to estimate the most valid method to rescale DCE predicted values using the model fit (R2), Akaike Information Criteria (AIC), root mean squared error (RMSE), and mean absolute error (MAE). A representative sample of 1376 Brazilian adults participated (DCE, 1152; cTTO, 211). The linear mapping without intercept (R2 = 96%; AIC, - 44; RMSE, 0.0803; MAE, - 0.0479) outperformed the non-linear without intercept (R2 = 98%; AIC, - 63; RMSE, 0.1385; MAE, - 0.1320). Utilities ranged from 1 (full health) to - 0.0059 (the worst health state). Highest weights were assigned to having pain or discomfort (pain/discomfort), followed by walking about (mobility), looking after myself (self-care), doing usual activities (usual activities), and feeling worried, sad, or unhappy (anxiety/depression). This study elicited the Brazilian EQ-5D-Y-3L value set using a mixed-logit DCE model with a power parameter based on a linear mapping without intercept, which can be used to estimate the quality-adjusted life-years for economic evaluations of health technologies targeting the Brazilian youth population.

Threshold mixed data sampling logit model with an application to forecasting US bank failures

Threshold mixed data sampling logit model with an application to forecasting US bank failures

The smallest worthwhile effect on pain intensity of exercise therapy for people with chronic low back pain: a discrete choice experiment study.

OBJECTIVE: To identify the smallest worthwhile effect (SWE) of exercise therapy for people with non-specific chronic low back pain (CLBP). DESIGN: Discrete choice experiment. METHODS: The SWE was estimated as the lowest reduction in pain that participants would consider exercising worthwhile, compared to not exercising i.e., effects due to natural history and other components (e.g., regression to the mean). We recruited English-speaking adults in Australia with non-specific CLBP to our online survey via email obtained from a registry of previous participants and advertisements on social media. We used discrete choice experiment to estimate the SWE of exercise compared to no exercise for pain intensity. We analysed the discrete choice experiment using a mixed logit model, and mitigated hypothetical bias through certainty calibration, with sensitivity analyses performed with different certainty calibration thresholds. RESULTS: Two-hundred and thirteen participants completed the survey. The mean age (±SD) was 50.7±16.5, median (IQR) pain duration 10 years (5-20), and mean pain intensity (±SD) was 5.8±2.3 on a 0-10 numerical rating scale. For people with CLBP the SWE of exercise was a between-group reduction in pain of 20%, compared to no exercise. In the sensitivity analyses, the SWE varied with different levels of certainty calibration; from 0% without certainty calibration to 60% with more extreme certainty calibration. CONCLUSION: This patient-informed threshold of clinical importance could guide the interpretation of findings from randomised trials and meta-analyses of exercise therapy compared to no exercise.

Quantifying the Impact of Risk Factors on Direct Compensation Property Damage in Canadian Automobile Insurance

This study presents a statistical analysis assessing the impact of various risk factors on direct compensation property damage (DCPD) claims in private passenger vehicle accidents. Using automobile insurance data in Ontario, Canada for the decade years period between 2003 and 2012, a statistical model of property damage was explored via a generalized linear binary logit mixed model and considered the imbalance between the classes of insureds. The results indicate that several risk factors have a significant impact on the likelihood of DCPD claims, including usage, training, outstanding loss, and incurred loss. The effects of these risk factors were observed under the weights — the number of trials used to generate each success proportion — in the different classes of insureds. The generalized linear mixed models (GLMMs) analysis provides a powerful tool for quantifying the impact of risk factors on binary outcomes, which are called DCPD claims and property damage (PD) claims covered by third-party liability (TPL) insurance. These models can also inform insurance underwriting and policy design, focusing on identifying the most significant risk factors. The performance metrics calculated by considering the class imbalance in binary outcomes verify the resulting model’s ability to accurately predict classes. The F1 score, an evaluation metric to measure the performance of classification, was calculated as 0.934. In addition, PR AUC, which is the area under the Precision-Recall (PR) curve, was computed as 0.953. These high scores indicate that the resulting model performs well in the classification. The other metrics also support the classification accuracy of this model. The findings of the analysis can help insurers better understand the underlying drivers of property damages and develop more accurate and effective strategies for risk mitigation. Furthermore, this study highlights the importance of developing class-specific risk assessment models to account for the imbalance across different classes.

The new frontier of urban mobility: a scenario-based analysis of autonomous vehicles adoption in urban transportation

ABSTRACT Urban mobility, as a critical component of urban development and sustainability, is on the brink of a major revolution with the advent and widespread adoption of autonomous vehicles (AVs). However, the ambitious vision of a transformative shared economic future and decreased car ownership conflicts with the notion that car dependency will be worsened, leading to significant uncertainties and potential challenges surrounding AVs, particularly in developing countries. In this context, this study contributes to the literature by addressing these uncertainties, through an empirical examination of preferences toward four distinct modes of transportation: “public transport”, “conventional private car”, “autonomous private car”, and “shared autonomous taxi”. To accomplish this, a scenario-based stated preference survey was administered in Istanbul, Turkey, as an example of a megacity in a developing country. The collected data were analyzed using discrete choice models (multinomial and random parameter (mixed) logit models) to reveal the preferences of the survey participants. The findings indicate that considering the social status and significance associated with private vehicle ownership, the challenges of popularizing shared and on-demand AV modes should not be underestimated.

Behavioral differences between young adults and elderly travelers concerning the crowding effect on public transit after the COVID-19 pandemic

After the COVID-19 pandemic, passengers' concerns about crowding on public transit vehicles has increased due to their fear of being infected. Since WHO has announced that the elderly are the most vulnerable to COVID-19, it is important to understand the impact of crowding on the elderly and provide policy implications on public transit vehicles. The aim of this study is to analyze the behavioral differences between young adults and the elderly concerning the crowded conditions on public transit with a stated preference survey in Seoul metropolitan area. Crowding multipliers are estimated to quantify the impedance on crowding based on the mixed logit model. The results indicate that the elderly are more sensitive to crowding on public transit vehicles than younger people, and subway passengers are affected more by crowding than passengers who are using other modes of transit. In addition, it is confirmed that the preference of the elderly for public transit varies according to gender, income, and the purpose of travel. These analysis results can provide implications for setting new policy directions to improve the public hygiene and comfort of elderly people when they use public transit in the post-COVID-19 world.

Do environmental awareness and food safety information nudges enhance youth’s affinity to safer food consumption? Findings from an experiment in Nepal

PurposeUnsafe food consumption results in adverse health conditions, foodborne illness and undernutrition among households and communities. The consumption of food contaminated with harmful microorganisms or with harmful pesticide residuals results in adverse health conditions and undernutrition. However, there are a number of challenges to maintaining food safety in the food systems of developing countries, like Nepal, where awareness of food safety is low and research on these issues is lacking.Design/methodology/approachWe conducted an experiment among youth aged between 20 and 26 years in Nepal to assess their food safety awareness and affinity to safer fresh produce choices. In the classroom setting experimentations with and without information nudges conducted among 224 youth participants, participants chose one fresh produce packet among the four. We analyzed results using multinomial and mixed logit models appropriate for discrete choice modeling.FindingsWe found that the youth’s perceived higher importance of sustainable food systems and their knowledge levels on microbial contamination and foodborne illnesses play significantly positive roles. The likelihood of choosing microbial safety-labeled fresh produce or both microbial- and chemical safety-labeled fresh produce increased with nudging among those who have some knowledge of microbial contamination and foodborne illnesses – we found that the interaction of nudging and level of knowledge is significantly positive. Youth belonging to higher income classes do not necessarily have a higher affinity to safer fresh produce but with nudging, the higher income class youth have a higher likelihood of choosing safer fresh produce choices.Research limitations/implicationsYouth engagement and their awareness of food safety could be one of the important strategies to potentially develop them as effective promoters, adopters and educators in enhancing food safety in food systems in Nepal. Our predicted premium for food safety attributes points to the potential scope for the emerging market segment or business opportunities augmenting food safety in Nepal.Originality/valueWe examined the factors influencing the safer fresh produce choices among youth in Nepal. We tested whether awareness levels of microbial contamination and foodborne illness and information nudging affect the likelihood of safer fresh produce choice. Then we predicted the willingness to pay (premium) for safety attributes. To the best of our knowledge, none of the previous studies have examined this aspect in Nepal.

The value of waiting spaces: Tourists’ willingness to pay for high-quality bus terminals

Transport quality has long been recognized as an important factor in influencing travelers’ behavior, and transport terminal quality undoubtedly plays no small part. Indeed, transit promotion policies explicitly based on qualitative factors and high-standard architecture are increasingly being adopted in designing new bus terminals and stops. High architectural standards can be found in several bus terminals worldwide. Nevertheless, the literature in transport sector has yet to explore the impact of bus terminal “hedonic quality” on users’ behavior or use their willingness to pay (WTP) in cost–benefit analyses or other transport policy applications. Hedonic value is here intended as the aggregate of all elements related to travelers’ pleasure in spending time in a terminal where architectural beauty and the passenger services offered are arguably the most visible and representative attributes. Within this context, we propose a quantitative analysis of the perceived hedonic value of bus nodes in terms of users’ WTP (pure preference) for a high-quality bus terminal. To this end a discrete choice experiment based on a visual immersive experience was performed, and data were collected from 324 residents of Milan (Italy) traveling for tourism. Different binomial Mixed Logit models with panel data and random coefficients were estimated for the purpose. Monetary valuation of pure preference for using a high-quality bus terminal was estimated with a mean of about 25 % of the average trip cost observed for the case study considered. This pure preference means that the Italian tourist is willing to spend up to €4.35/trip more for a high-quality bus waiting space or travel up to 28.2 min/trip more, instead of using a traditional bus terminal for the same trip. Category-specific analyses show that females > 30 years old have a greater pure preference for a high-quality bus terminal (up to + 220 %) against males and young users. The employed also have a higher (+42 %) pure preference for the beauty of a bus terminal against others. The results of this paper should be compared with those from other case studies as they have a potential impact for transportation planning applications, underling the importance of also incorporating “hedonic quality” as explicit design variables for new/revamped transportation hub. At the same time, new challenges are posed for modeling user behavior and determining quality-related indicators and measures.

How should regulatory schemes be optimized to enhance deterrence against medical insurance fraud by enrollees? Evidence from a discrete choice experiment in China

Medical insurance fraud (MIF) poses a substantial global financial challenge, necessitating effective regulatory strategies, especially in China, where such measures are in a critical developmental phase. This study investigates the effectiveness of various regulatory components in deterring MIF among enrollees and explores preference heterogeneity among individuals with different characteristics, utilizing a discrete choice experiment survey. Grounded in deterrence theory, our conceptual framework incorporates five attributes: intensity of economic penalties, restrictions on medical insurance benefits, deterioration of social reputation, and certainty and celerity of penalties. Employing a D-efficiency design, 24 choice sets were generated and blocked into three versions. A multistage stratified sampling method was adopted to collect data from the basic medical insurance enrollees in Shanghai. The survey was conducted from September to October 2022. The sample representativeness was further improved via the entropy balancing approach. Data from the final sample of 1034 respondents were analyzed using mixed logit models (MIXLs), incorporating interactions with individual characteristics to assess preference heterogeneity. Results reveal that escalating economic penalties, suspending insurance benefits, listing individuals as unfaithful parties, ensuring penalty certainty, and expediting enforcement significantly enhance the deterrent effect. We observed preference heterogeneity across different demographics, including age, gender, education, health status, and employment status. The study underscores the pivotal role of economic penalties in deterring MIF, while also acknowledging the significance of non-economic measures such as enforcement efficiency and social sanctions. These findings offer valuable insights for policymakers to tailor and strengthen regulatory schemes against MIF, contributing to the advancement of more effective and precise healthcare policies.

The effect of freeway toll pricing on travel mode changes, route changes, and departure time changes.

This study critically examines the influence of freeway toll pricing on the travel behavior of intercity commuters in Iran, focusing on mode choice, route selection, and departure times. A survey of 921 commuters was conducted to gather data, which was then analyzed using a mixed logit model. The key findings reveal that a one percent increase in toll prices leads to a 1.5101 percent decrease in the likelihood of commuters maintaining their usual travel mode. Conversely, the probability of selecting alternative travel options rises by 2.5129 percent. A notable change is observed in route choice, with a 1000 toman increase in tolls raising the likelihood of route change by 0.320 percent. Furthermore, increased tolls are associated with a higher probability of changes in travel mode (0.1188%) and departure time (0.1078%). This research highlights the significant behavioral shifts among commuters in response to toll pricing adjustments and underscores the need for strategic toll management in transportation planning.

A study on diversion behavior in weaving segments: Individualized traffic conflict prediction and causal mechanism analysis

Lane change behavior disrupts traffic flow and increases the potential for traffic conflicts, especially on expressway weaving segments. Focusing on the diversion process, this study incorporating individual driving patterns into conflict prediction and causation analysis can help develop individualized intervention measures to avoid risky diversion behaviors. First, to minimize measurement errors, this study introduces a lane line reconstruction method. Second, several unsupervised clustering methods, including k-means, agglomerative clustering, gaussian mixture, and spectral clustering, are applied to explore diversion patterns. Moreover, machine learning methods, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Attention-based LSTM, eXtreme Gradient Boosting (XGB), Support Vector Machine (SVM), and Multilayer Perceptron (MLP), are employed for real-time traffic conflict prediction. Finally, mixed logit models are developed using pre-conflict condition data to investigate the causal mechanisms of traffic conflicts. The results indicate that the K-means algorithm with four clusters exhibits the highest Calinski-Harabasz and Silhouette scores and the lowest Davies-Bouldin scores. With superior classification accuracy and generalization ability, the LSTM is used to develop the personalized traffic conflict prediction model. Sensitivity analysis indicates that incorporating the diversion patterns into the LSTM model results in an improvement of 3.64% in Accuracy, 7.15% in Precision, and 1.34% in Recall. Results from the four mixed logit models indicate significant differences in factors contributing to traffic conflicts within each diversion pattern. For instance, increasing the speed difference between the target vehicle and the right preceding vehicle benefits traffic conflict during acceleration diversions but decreases the likelihood of traffic conflicts during deceleration diversions. These results can help traffic engineers propose individualized solutions to reduce unsafe diversion behavior.

Estimating Substitution Patterns and Demand Curvature in Discrete-Choice Models of Product Differentiation

Abstract We extend BLP's aggregate discrete-choice model of product differentiation to create more flexibility in the price functional form. We apply a Box-Cox specification, which relaxes the typical unit demand assumption and creates flexibility on demand curvature. The model provides a unifying framework for mixed logit and mixed CES models, while remaining computationally tractable. We provide an illustrative application to the ready-to-eat cereals market. This shows that the cross-sectional relation between price elasticities and average prices per product is more in line with descriptive elasticity patterns, and that substitution between product pairs may be affected to some extent.