Based on the results of the regression analyses conducted in the identified studies the attribute “Care provider” was the most important in 4 of the DCEs [17, 18, 25, 28]. Caldow et al.  for example show that it is most important for respondents to see a GP rather than a practice nurse. This is even more important than “Continuity of health professional”, “Waiting time till appointment”, “Likelihood of having ill- ness cured” and “Length of consultation”. 3 Furthermore, 3 studies identified “Shared decision making” as being most important to the respective participants [19, 20, 30]. Hjelmgren and Anell  state that having influence on the decision of the care they receive is most important for respondents, followed by “Individual choice of GP/care provider”, “Waiting time till appointment” and “Price”. Other studies show that “Waiting time” is the most im- portant attribute when choosing a primary care alternative [3, 22, 29, 32]. In their study Pedersen et al.  identify the typical waiting time until appointment (for routine tasks) to be most important for patients, for instance. This attribute is more important than “Distance to the prac- tice”, “Waiting time on the telephone” and “Length of consultation”. The other discretechoiceexperiments de- termine “Information and explanation (on medicines/ treatment/problem)” [4, 27], “Continuity of health professional/physician’s knowledge of the patient” [24, 28], “Receiving he ‘best’ treatment” [26, 33], “Thoroughness of physical examination”  and “Individual choice of GP/care provider”  as the most important attribute.
environmental quality improvements not captured by the attributes that should be accounted for in welfare estimation. That is, if a respondent decides to choose only from the policy alternatives, and is considering at least cost to arrive at her/his final choice, it can be argued that ASCs should be included. In contrast to the ECLC models that incorporate AN-A behaviour, the MXL models do not allow this distinction, and the resulting contribution of the ASCs to the CS estimates is particularly large, especially for the SRB sample. The CS estimates based on the MXL models should therefore be treated with caution. Finally, if ASCs are included, it is implicitly assumed that the two river basins do not differ in unobserved aspects (i.e., in aspects that are not covered by the attributes that are related to water scarcity and ecological status). If this is the case, Morrison et al. (2002) suggest that it may be prudent to only use MWTP estimates for benefit transfer. While the two river basins share similar characteristics regarding the challenges they face, we cannot deny the possibility that unobserved differences between sites may exist, for example related to differences in patterns of water use by the respective populations. Irrespective of the above issues, our main finding that econometric specification error related to AN-A behaviour can matter in function benefit transfer based on discretechoiceexperiments is not affected by whether or not we include ASCs for welfare estimation.
Discretechoiceexperiments (DCEs) are the most com- mon type of ordinal task used in health services research to estimate utilities based on patient choices. DCEs can estimate health state values for two reasons. First, they are compatible with Lancaster's theory of value, which states that the total utility of a state can be decomposed into util- ities of characteristics that describe it ; given an appro- priate statistical design , utilities of alternative health states (profiles/specifications) allow one to infer utilities of attribute levels that describe them. Second, they are consistent with random utility theory (RUT), a well-tested theory of human decision-making [7,8]. RUT assumes that the total utility of a good/service can be expressed as the sum of two components, one fixed (systematic), and a second random (stochastic). If the random component is an independently and identically distributed (iid) extreme value type 1 (Gumbel) random variate, then the underlying choice process is consistent with McFadden's (1974) conditional (multinomial) logit model, and this model can be used to estimate the elements of the fixed component . That is, the relative choice frequencies reveal the individuals' preferences (utilities), which can be estimated from the frequencies as a function of attribute levels. When the above model holds, and in particular when no choice probabilities equal 0 or 1, we refer to the model as the conventional random utility model.
Results: Twenty-seven studies were included, with over half set in low- and middle-income countries. There were more studies published in the last four years than the previous ten years. Doctors or medical students were the most studied cadre. Studies frequently pooled results from heterogeneous subgroups or extrapolated these results to the general population. Only one third of studies included an opt-out option, despite all health workers having the option to exit the labour market. Just five studies combined results with cost data to assess the cost effectiveness of various policy options. Comparison of results from similar studies broadly showed the importance of bonus payments and postgraduate training opportunities and the unpopularity of time commitments for the uptake of rural posts. Conclusions: This is the first systematic review of discretechoiceexperiments in human resources for health. We identified specific issues relating to this application of which practitioners should be aware to ensure robust results. In particular, there is a need for more defined target populations and increased synthesis with cost data. Research on a wider range of health workers and the generalisability of results would be welcome to better inform policy.
There are numerous implementation strategies available that have proven to be at least moderately effective in bringing about change [1-3]. Current insight in imple- mentation research is that the choice of implementation strategy should be guided by a diagnostic analysis that starts with describing the gap between current care and optimal care . An important part of the diagnostic anal- ysis is the identification of barriers and facilitators to change. Until now, studies identifying barriers and facili- tators to change have been carried out using a combina- tion of qualitative and quantitative methods, such as case- specific questionnaires, semi-structured in-depth inter- views, focus group interviews, and non-participating observation . These methods have some limitations. First, they generally yield many barriers, but do not pro- vide information with respect to the relative importance, or prioritizing, of the barriers. Second, existing instru- ments hardly differentiate between barriers, and conse- quently may overestimate the importance of less important barriers and underestimate the importance of more important barriers. Third, these traditional methods have a non-compensatory character, which may be prob- lematic because in decision processes concerning the implementation of change it is often the case that facilita- tors can partly compensate for barriers. These limitations associated with the methods that are currently applied in implementation research have revealed the need for an alternative research methodology for the evaluation of barriers and facilitators. Discretechoiceexperiments (DCE) to investigate preferences are relatively new in the health care sector, and may be of value in the field of implementation research. DCE is a stated preference method that presents individuals with a number of choices. Each choice consists of two or more hypothetical profiles, and for each choice, people are asked which pro- file they would choose. Forcing people to make choices and trade-offs is a big advantage of DCE over the methods that are currently applied in implementation research. For us, this is a strong motivation to introduce DCE in imple- mentation research. If DCE proves to have a complemen- tary value for the evaluation of barriers and facilitators, the choice of implementation strategy will be based on factors that more accurately reflect individuals' prefer-
The HIV burden remains highly concentrated in southern Africa, with an estimated adult prevalence of 8.8% in Malawi and 13% in Zambia between 2014 and 2015 (1, 2). Despite substantial progress, there is a significant gap in HIV testing, with just 73% and 67% of HIV- positive individuals know their status in Malawi and Zambia, respectively (3). HIV self-testing (HIVST), now recommended by the World Health Organisation, is defined as the process by which a person collects his/her own specimen, performs a test for HIV and interprets the results (4). A reactive HIVST needs to be confirmed by a healthcare professional with referral to ART services if HIV positive. HIVST has demonstrated high acceptability, though rates of linkage to confirmatory HIV testing and treatment have remained sub-optimal among self-testers (5, 6). Discretechoiceexperiments (DCE) are a valuable way of measuring and quantifying user preferences for goods and services, particularly when there is a dearth of data around observed behaviour (7-9) and only limited service configurations are available. DCE have been used in a myriad of health interventions in sub-Saharan Africa including the investigation of populations’ preferences for the design of sexual and reproductive health care services (10-12), including to inform the design of a voluntary medical circumcision program (13) and guide of HIVST kit distribution (14). This research uses a DCE to identify drivers of demand for linkage into confirmatory testing and care following a reactive HIVST in Malawi and Zambia.
Discretechoiceexperiments have become increasingly used in health services research, but primarily to assess patient-stated preferences and willingness to pay for dif- ferent models of health care service delivery [47-50]. There are still only a small number of studies that have used this methodology to analyse the job preferences of health care providers. The aim of this article is to review the existing literature on the use of discretechoice experi- ments to study HR issues in both developed and develop- ing countries. The intention is to draw lessons on the value of this relatively new methodology to inform HR policy development in developing countries. This paper first introduces the basic principles of DCE methods, then the methodology of our literature review is described. The main part of the paper describes the DCE studies we iden- tified and summarizes their findings. The final discussion focuses on some cross-cutting lessons as well as the advan- tages and limitations of DCE methods for HR research. Methods
these approaches provides concrete guidance for imple- menters on which combination of characteristics will achieve the greatest uptake. Some studies have used discretechoiceexperiments (DCEs) to inform programme design and identify potential barriers and facilitators of uptake [19–21]. DCEs are a survey-based approach to eliciting user preferences. They allow the es- timation of user values in the absence of observable markets, where services are provided for free or have not yet been introduced. They can measure the strength of preferences between service attributes, for example, valuing waiting times, prices and provider gender, inde- pendently. Lastly, they can identify where preferences differ between individuals, which is particularly useful when complex interventions include targeting specific user groups.
In addition, those previous studies did not examine willingness to pay for the current CBHI schemes by providing different alternative using a discretechoice experiment (DCE). In this time accepted eliciting stated preference model in health economics is discretechoiceexperiments (DCE). Discretechoiceexperiments (DCEs) involve a process of developing, testing and optimizing the experiment questionnaire [9, 14-16]. DCEs need respondents select their choice over sets of supposed options which each alternative explained by a set of attributes, and all attribute catches one of a number of levels [9, 14-16].
Materials and methods: Subjects included 100 patients with arthritis and 60 board-certified orthopedic surgeon physicians in South Korea. Through a systematic review of the literature, beneficial attributes of using Cox-2 inhibitors were defined as a decrease in the Western Ontario and McMaster Universities Arthritis Index for pain score and improvement in physical function. Likewise, risk attributes included upper gastrointestinal (GI) complications and cardiovascular (CV) adverse events. Discretechoiceexperiments were used to determine preferences for these four attributes among Korean patients and physicians. Relative importance and maximum acceptable risk for improving beneficial attributes were assessed by analyzing the results of the discretechoice experiment by using a conditional logit model.
5.7 Figure 8 illustrates how the magnitude of the error term ǫ influences the simulation results. The standard devia- tion of the error term is set in relation to the average ∆ V (see Figure 5) and is varied between 40% and 400% of the average ∆ V . The observed variable is again the fitness variable described above, the ratio between round- wood sold and the annual allowable cut, which should always be close to one. It can be observed that the variance of the curves decreases with an increasing standard deviation of the error term. The reason for this eﬀect is in the error term whereby increasing its standard deviation respectively increases the randomization of the entire utility function. This means that utility functions which previously resulted in very low market ac- tivity now have a higher probability of allowing normal market participation. However, this leads to the eﬀect that more negotiation rounds are necessary when the standard deviation increases. The additional negotiation rounds compensate for the increased randomness in the utility function; if the randomness is high, the prob- ability that lucrative oﬀers are rejected and unprofitable ones accepted increases. Therefore, a high standard deviation of the error term also leads to lower cost-eﬀectiveness of the market. However, including the error term in the utility functions is important, as otherwise they are no longer consistent with random utility theory (cf. section DiscreteChoiceExperiments; Louviere et al. (2010)).
The estimation of marginal utility of income in discretechoiceexperiments is of crucial importance for the estimation of willingness to pay (WTP) and welfare estimates. Despite this central importance, there are only few investigations into the impact of the design of the cost attribute vector on choices and WTP estimates. We present a conceptual framework that describes why cost vector effects might occur in choiceexperiments, and investigate cost vector effects empirically drawing on data from a choice experiment in the context of peatland restoration in Scotland. This study employs a split sample approach with three different cost vectors that vary considerably in the cost levels offered to respondents, and investigates differences between treatments with respect to marginal WTP estimates, sta- tus quo choice, use of systematic decision strategies and attribute non-attendance. A key ﬁnding is that the choice of cost vectors can affect the incidence of decision strategies. After accounting for the differential use of a decision strategy that might not be consistent with random utility modelling, cost vectors that are higher in magnitude result in higher WTP, in line with an anchoring hypothesis. We ﬁnd weak support that marginal WTP of lower income respondents is affected differently compared to higher income respondents through the use of different cost vectors. Differences in welfare estimates resulting from the use of different cost vectors might change outcomes of cost-beneﬁt analyses. We therefore recommend that researchers include tests of sensitivity of welfare estimates to different cost vectors in their study design.
Discretechoiceexperiments are a commonly used tech- nique to address a range of important healthcare ques- tions. DCEs constitute an attribute-based measure of benefit, with the assumptions that first, healthcare inter- ventions, services or policies can be described by their at- tributes or characteristics and second, the levels of these attributes drive an individual ’ s valuation of the healthcare good. Within a DCE, respondents are asked to choose be- tween two or more alternatives. The resulting choices re- veal an underlying utility function (i.e., an economic measure of preferences over a given set of goods or ser- vices). The approach combines econometric analysis with experimental design theory, consumer theory, and ran- dom utility theory, which posits that consumers generally choose what they prefer, and where they do not, this can be explained by random factors [6, 8, 9]. Meanwhile, con- joint analysis originated in psychology to address the mathematical representation of the behavior of rankings observed as an outcome of systematic, factorial manipula- tion of multiple measures. Although there is a distinction between conjoint analysis and DCE, the two terms are used interchangeability by many researchers .
should be adjusted for ANA. The present study was con- ducted alongside the pilot for a DCE and varied the number of attributes across choice sets to identify the effect of complexity on ANA. As a consequence, we observed limited variation across attribute levels for some attributes and could not account for the effect of all attributes when estimating utility functions. Lagarde  found that whilst willingness-to-pay estimates were sensitive to ANA, the behavioural prediction of DCE models was not affected by ANA. One explanation for this may be that consumers are so accustomed to using heuristics or decision rules in complex or uncertain situ- ations that they are well practised in seeking out infor- mation that will be useful to them in their final decision (in essence, conferring zero utility for any attributes superfluous to their needs). Thus, reading attribute and alternative labels may be sufficient for some consumers to decide if the subsequent information available is worthwhile attending to or not.
The six attributes and corresponding levels described in Table 1 give rise to a possible combination of 729 (3 6 ) treatment outcome scenarios. As this full factorial design of possible scenarios was not feasible to present to each patient, a sample was generated using a fractional factor- ial experimental design. An efficient design was gener- ated using NGENE software, such that attributes were varied independently from one another across the sce- narios and that standard errors were minimised. This model allows for multiple versions of the questionnaire, reducing the burden on patients and increasing the stat- istical efficiency of the study. Two versions of the DCE questionnaire were generated, each containing six differ- ent choice sets. Each DCE questionnaire had the same number of scenarios with the same attributes, but differ- ent attribute levels in each questionnaire. The sample of scenarios were then organised into pairwise profiles la- belled ‘Choice A ’ and ‘Choice B’ , and participants were asked to choose between the pair of choices in each of the six choice sets. Furthermore, an opt-out option was included at the end of each choice set where patients were asked if, given the scenarios presented, they would still have the operation or prefer to remain in their current health state. This opt-out option was included as it reflects the voluntary nature of elective TKA in real life. For an example of a discretechoice task see Table 2.
possibility that the data would contain systematic error introduced by the challenge of completing the DCE. Rather than simply eliminate data of low quality (and potentially introduce bias due to the choice of quality criteria), we chose first to include all data in the primary analysis, second to conduct a sensitivity analysis in which data with high risk of low quality were excluded, and third to conduct the latent class analysis which identified a group of patents with inconsistent and weak preferences which included many of the individuals who met criteria for low-quality data. While our findings make it clear that some aspects of person-centred care do matter more than others, they act as a starting point for further enquiry including where sufficiency thresholds lie (as described above, we compared enhanced attributes against neutral, not negative ones); whether particular combinations of features are important; and whether preferences change over the course of illness or in different healthcare contexts. These may require different study designs and interpretive approaches.
About 25% of the respondents from the screened popula- tion was excluded due to .10% missing choice tasks (ie, missing answers on two or more choice tasks). In most instances these were consecutive choice tasks; therefore, it is assumed that this was most likely due to accidently skip- ping a page (implying 2–4 missed choice tasks at once). To ensure robustness of our results, all DCE analyses were conducted separately for data sets including and excluding these respondents. In addition, since older age and male gender are associated with a higher likelihood of CRC and thereby positive FIT, 25 significant differences in demographic
The results are remarkable because of their consistency across the countries. By far the strongest predictors of job choice were shown to be access to continuing pro- fessional development and the presence of functioning human resources managemant. Consistent with similar works we find pay and allowances to be important and significantly positively related to utility, but financial re- wards are not as fundamental a factor underlying em- ployment preferences as many may have previously believed. There is evidence to indicate diminishing mar- ginal utility in relation to pay in the three countries. Loca- tion (urban vs rural) had the smallest effect on utility for job choice in all three countries. These findings are im- portant in the context where efforts to address the human resources crisis have focused primarily on increasing salar- ies and incentives, as well as providing additional allow- ances to work in rural areas. Our conclusion is that improving human resources management, and in particu- lar access to continuing professional development, may prove a more effective motivation and retention strategy. Appendix 1
We used a binary dependent variable model in order to force a choice. This is because, in reality, transplant decisions have to be made and medical professionals face a forced choice when allocating kidneys because of donor scarcity. Moreover, pilot interviews revealed that many respondents felt uncomfortable with deciding who to transplant. Therefore, it was judged that including a „cannot decide‟ option might have triggered such a response from people who in reality were not indifferent. An alternative would have been to allow choices between more than 2 potential recipients using a multinomial model, or to have more attributes and levels; but this would have complicated decision making . Moreover, since many renal patients suffer from fatigue we wanted to avoid complicated decisions, because when complexity increases there is evidence that respondents may be more inclined to use simplifying heuristics  compromising response reliability. Copies of the final DCE questionnaires are provided in Additional file 2.