Data from: How to use discrete choice experiments to capture stakeholder preferences in social work research
Data files
May 27, 2024 version files 25.58 MB
-
03_table_s2.csv
-
04_design_wide_cost_dummy.csv
-
05_design_long_cost_continuous.csv
-
08_population_data.csv
-
09_population_parameters.csv
-
12_choice_data.csv
-
14_results_cl.csv
-
15_results_rpl.csv
-
17_sample_size_rpl_results.csv
-
23_wtp_results_cl.csv
-
24_wtp_results_rpl.csv
-
README.md
Abstract
The primary article (cited below under "Related works") introduces social work researchers to discrete choice experiments (DCEs) for studying stakeholder preferences. The article includes an online supplement with a worked example demonstrating DCE design and analysis with realistic simulated data. The worked example focuses on caregivers' priorities in choosing treatment for children with attention deficit hyperactivity disorder. This dataset includes the scripts (and, in some cases, Excel files) that we used to identify appropriate experimental designs, simulate population and sample data, estimate sample size requirements for the multinomial logit (MNL, also known as conditional logit) and random parameter logit (RPL) models, estimate parameters using the MNL and RPL models, and analyze attribute importance, willingness to pay, and predicted uptake. It also includes the associated data files (experimental designs, data generation parameters, simulated population data and parameters, simulated choice data, MNL and RPL results, RPL sample size simulation results, and willingness-to-pay results) and images. The data could easily be analyzed using other software, and the code could easily be adapted to analyze other data. Because this dataset contains only simulated data, we are not aware of any legal or ethical considerations.
README: Data from: How to Use Discrete Choice Experiments to Capture Stakeholder Preferences in Social Work Research
This dataset supports the worked example in:
Ellis, A. R., Cryer-Coupet, Q. R., Weller, B. E., Howard, K., Raghunandan, R., & Thomas, K. C. (2024). How to use discrete choice experiments to capture stakeholder preferences in social work research. Journal of the Society for Social Work and Research. Advance online publication. https://doi.org/10.1086/731310
The referenced article introduces social work researchers to discrete choice experiments (DCEs) for studying stakeholder preferences. In a DCE, researchers ask participants to complete a series of choice tasks: hypothetical situations in which each participant is presented with alternative scenarios and selects one or more. For example, social work researchers may want to know how parents and other caregivers prioritize different aspects of mental health treatment when choosing services for their children. A DCE can explore this question by presenting scenarios that include different types of mental health care providers, treatment methods, costs, locations, and so on. Caregivers’ stated choices in these scenarios can provide a lot of information about their priorities.
In a DCE, the scenarios presented to participants are called alternatives. The characteristics of the alternatives are called attributes, and the attribute values presented to participants are called levels. Participants' choices are typically analyzed according to random utility theory, which holds that utility is a latent variable underlying people's observed choices. Utility has a systematic component derived from the attributes of alternatives and the individual characteristics of participants, as well as a random component.
Our article includes an online supplement with a worked example demonstrating DCE design and analysis with realistic simulated data. The worked example focuses on caregivers' priorities in choosing treatment for children with attention deficit hyperactivity disorder. This submission includes the scripts (and, in some cases, Excel files) that we used to create and analyze the worked example, as well as the associated data files and images.
In the worked example, we used simulated data to examine caregiver preferences for 7 treatment attributes (medication administration, therapy location, school accommodation, caregiver behavior training, provider communication, provider specialty, and monthly out-of-pocket costs) identified by dosReis and colleagues (2007, 2015) in a previous DCE. We employed an orthogonal design (so named because all pairs of attribute levels appear with equal frequency, making the attributes independent of each other) with 18 choice tasks. Our design included 1 continuous attribute (cost) and 12 categorical attributes, which were dummy-coded. The online supplement shows all of the attribute levels and which combinations of levels were presented in the simulated DCE.
Using the parameter estimates published by dosReis et al., with slight adaptations, we simulated utility values for a population (N=100,000), then selected a sample (n=500) for analysis. Relying on random utility theory, we used the multinomial logit (MNL, also known as conditional logit) and random parameter logit (RPL) models to estimate utility parameters. The MNL model assumes that the systematic component of utility is distributed uniformly across individuals. The RPL model assumes that utility varies randomly across individuals according to a specified distribution.
In addition to estimating utility, we measured the relative importance of each attribute, estimated caregivers’ willingness to pay (WTP) for differences in attributes (e.g., how much they would be willing to pay for their child to see one type of provider versus another) with bootstrapped 95% confidence intervals, and predicted the uptake of three treatment packages with different sets of attributes. This dataset includes both the simulated source data and the processed results. The online supplement of the referenced article describes the methods in greater detail.
The next section of this document describes how the files in this submission relate to each other, and the two sections following that describe (a) the structure and content of the data files and (b) the structure and content of the Excel files, which contain mixtures of data and formulas. The code included with this submission can be used to replicate the data. Before the R code is run, the working directory must be set (near the top of each script, where it says "set your working directory here").
File structure
The following table describes the files in the order in which they were created. It also describes how they relate to each other.
File | Description | Source Program or Input File(s) | Output File(s) |
---|---|---|---|
00_README.md | this file | ||
01_design.r | R code to identify an appropriate experimental design | 03_table_s2.csv, 04_design_wide_cost_dummy.csv, 05_design_long_cost_continuous.csv | |
02_design.sas | SAS code to identify an alternate design | ||
03_table_s2.csv | experimental design (shown in Table S2 of the online supplement) | 01_design.r | |
04_design_wide_cost_dummy.csv | experimental design, dummy-coded | 01_design.r | |
05_design_long_cost_continuous.csv | experimental design, adjusted to make cost continuous | 01_design.r | |
06_population_simulation_input.xlsx | parameters used for population simulation | adapted from dos Reis et al. (2017) | |
07_simulate_population.r | R code to simulate the study population | 06_population_simulation_input.xlsx | 08_population_data.csv, 09_population_parameters.csv |
08_population_data.csv | simulated population data | 07_simulate_population.r | |
09_population_parameters.csv | true values of the population parameters | 07_simulate_population.r | |
10_sample_size_cl.r | R code to estimate required sample size for multinomial logit (conditional logit) model | 05_design_long_cost_continuous.csv, 09_population_parameters.csv | |
11_simulate_sample.r | R code to select sample and simulate choice data | 05_design_long_cost_continuous.csv, 08_population_data.csv | 12_choice_data.csv |
12_choice_data.csv | choice data from the study sample | 11_simulate_sample.r | |
13_analysis.r | R code to run multinomial logit (conditional logit) and random parameter logit models | 12_choice_data.csv | 14_results_cl.csv, 15_results_rpl.csv |
14_results_cl.csv | multinomial logit (conditional logit) results | 13_analysis.r | |
15_results_rpl.csv | random parameter logit results | 13_analysis.r | |
16_sample_size_rpl_simulation.r | simulation to estimate sample size requirements for random parameter logit model | 05_design_long_cost_continuous.csv, 08_population_data.csv, 14_results_cl.csv | 17_sample_size_rpl_results.csv |
17_sample_size_rpl_results.csv | sample size simulation results | 16_sample_size_rpl_simulation.r | |
18_sample_size_rpl_analysis.r | R code to analyze sample size simulation results for random parameter logit model | 09_population_parameters.csv, 17_sample_size_rpl_results.csv | |
19_attribute_importance_analysis.xlsx | Excel file to analyze attribute importance based on attribute level range and parameter estimates | 14_results_cl.csv, 15_results_rpl.csv | |
20_attribute_importance_plot.r | R code to plot attribute importance results | 19_attribute_importance_analysis.xlsx | 21_attribute_importance.png |
21_attribute_importance.png | plot of attribute importance results | 20_attribute_importance_plot.r | |
22_wtp_analysis.r | R code to estimate willingness to pay with bootstrapped estimates and confidence intervals | 12_choice_data.csv, 14_results_cl.csv | 23_wtp_results_cl.csv, 24_wtp_results_rpl.csv |
23_wtp_results_cl.csv | willingness-to-pay estimates for multinomial logit (conditional logit) model | 22_wtp_analysis.r | |
24_wtp_results_rpl.csv | willingness-to-pay estimates for random parameter logit model | 22_wtp_analysis.r | |
25_wtp_summary.r | R code to summarize willingness-to-pay results | 23_wtp_results_cl.csv, 24_wtp_results_rpl.csv | 26_wtp_plot.png |
26_wtp_plot.png | willingness-to-pay plot | 25_wtp_summary.r | |
27_predicted_uptake.xlsx | Excel file to predict uptake of specific alternatives | 14_results_cl.csv |
Description of data file contents
This section describes the contents of each individual data file, in the order in which the files are listed above.
03_table_s2.csv
experimental design (shown in Table S2 of the online supplement)
Each row in the file represents one of the 18 choice tasks. Each choice task had two alternatives. In the original study by dosReis et al., each attribute (including cost) had 3 levels (shown in the online supplement of the paper referenced above). In this file, each attribute level is coded as 1, 2, or 3.
Column | Description |
---|---|
meds1 | level of the medication administration attribute for alternative 1 |
therapy1 | level of the therapy location attribute for alternative 1 |
accommodation1 | level of the school accommodation attribute for alternative 1 |
cgtraining1 | level of the caregiver behavior training attribute for alternative 1 |
prvcomm1 | level of the provider communication attribute for alternative 1 |
prvspec1 | level of the provider specialty attribute for alternative 1 |
cost1 | level of the monthly out-of-pocket costs attribute for alternative 1 |
meds2 | level of the medication administration attribute for alternative 2 |
therapy2 | level of the therapy location attribute for alternative 2 |
accommodation2 | level of the school accommodation attribute for alternative 2 |
cgtraining2 | level of the caregiver behavior training attribute for alternative 2 |
prvcomm2 | level of the provider communication attribute for alternative 2 |
prvspec2 | level of the provider specialty attribute for alternative 2 |
cost2 | level of the monthly out-of-pocket costs attribute for alternative 2 |
04_design_wide_cost_dummy.csv
experimental design, dummy-coded
Each row in the file represents one of the 18 choice tasks. Each choice task had two alternatives. In the original study by dosReis et al., each attribute (including cost) had 3 levels (shown in the online supplement of the paper referenced above). In this file, each attribute level is coded using two dummy variables for levels 2 and 3, with level 1 as the reference category.
Column | Description |
---|---|
meds1_2 | =1 if medication administration = 2 for alternative 1, 0 otherwise |
meds1_3 | =1 if medication administration = 3 for alternative 1, 0 otherwise |
therapy1_2 | =1 if therapy location = 2 for alternative 1, 0 otherwise |
therapy1_3 | =1 if therapy location = 3 for alternative 1, 0 otherwise |
accommodation1_2 | =1 if school accommodation = 2 for alternative 1, 0 otherwise |
accommodation1_3 | =1 if school accommodation = 3 for alternative 1, 0 otherwise |
cgtraining1_2 | =1 if caregiver behavior training = 2 for alternative 1, 0 otherwise |
cgtraining1_3 | =1 if caregiver behavior training = 3 for alternative 1, 0 otherwise |
prvcomm1_2 | =1 if provider communication = 2 for alternative 1, 0 otherwise |
prvcomm1_3 | =1 if provider communication = 3 for alternative 1, 0 otherwise |
prvspec1_2 | =1 if provider specialty = 2 for alternative 1, 0 otherwise |
prvspec1_3 | =1 if provider specialty = 3 for alternative 1, 0 otherwise |
cost1_2 | =1 if monthly out-of-pocket costs = 2 for alternative 1, 0 otherwise |
cost1_3 | =1 if monthly out-of-pocket costs = 3 for alternative 1, 0 otherwise |
meds2_2 | =1 if medication administration = 2 for alternative 2, 0 otherwise |
meds2_3 | =1 if medication administration = 3 for alternative 2, 0 otherwise |
therapy2_2 | =1 if therapy location = 2 for alternative 2, 0 otherwise |
therapy2_3 | =1 if therapy location = 3 for alternative 2, 0 otherwise |
accommodation2_2 | =1 if school accommodation = 2 for alternative 2, 0 otherwise |
accommodation2_3 | =1 if school accommodation = 3 for alternative 2, 0 otherwise |
cgtraining2_2 | =1 if caregiver behavior training = 2 for alternative 2, 0 otherwise |
cgtraining2_3 | =1 if caregiver behavior training = 3 for alternative 2, 0 otherwise |
prvcomm2_2 | =1 if provider communication = 2 for alternative 2, 0 otherwise |
prvcomm2_3 | =1 if provider communication = 3 for alternative 2, 0 otherwise |
prvspec2_2 | =1 if provider specialty = 2 for alternative 2, 0 otherwise |
prvspec2_3 | =1 if provider specialty = 3 for alternative 2, 0 otherwise |
cost2_2 | =1 if monthly out-of-pocket costs = 2 for alternative 2, 0 otherwise |
cost2_3 | =1 if monthly out-of-pocket costs = 3 for alternative 2, 0 otherwise |
05_design_long_cost_continuous.csv
experimental design, adjusted to make cost continuous
Each pair of rows in the file represents one of the 18 choice tasks, with one row for each alternative presented in the task. Each choice task had two alternatives. In the original study by dosReis et al., each attribute (including cost) had 3 levels (shown in the online supplement of the paper referenced above). In this file, the cost variable is coded as continuous. Otherwise, each attribute level is coded using two dummy variables for levels 2 and 3, with level 1 as the reference category.
Column | Description |
---|---|
task | index of the choice task (1 to 18) |
alt | index of the alternative within the choice task (1 or 2) |
meds_2 | =1 if medication administration = 2, 0 otherwise |
meds_3 | =1 if medication administration = 3, 0 otherwise |
therapy_2 | =1 if therapy location = 2, 0 otherwise |
therapy_3 | =1 if therapy location = 3, 0 otherwise |
accommodation_2 | =1 if school accommodation = 2, 0 otherwise |
accommodation_3 | =1 if school accommodation = 3, 0 otherwise |
cgtraining_2 | =1 if caregiver behavior training = 2, 0 otherwise |
cgtraining_3 | =1 if caregiver behavior training = 3, 0 otherwise |
prvcomm_2 | =1 if provider communication = 2, 0 otherwise |
prvcomm_3 | =1 if provider communication = 3, 0 otherwise |
prvspec_2 | =1 if provider specialty = 2, 0 otherwise |
prvspec_3 | =1 if provider specialty = 3, 0 otherwise |
cost | monthly out-of-pocket costs in hundreds of US dollars |
08_population_data.csv
simulated population data
This file contains the population data that we simulated. Each row represents an individual. There are two columns for each categorical attribute and one column for cost. The number in each cell is the individual's utility for the specified attribute level, or, in the case of cost, for a $100 increase in cost.
Column | Description |
---|---|
id | ID of the individual (1 to 100,000) |
meds_2 | utility for medication administration = 2 |
meds_3 | utility for medication administration = 3 |
therapy_2 | utility for therapy location = 2 |
therapy_3 | utility for therapy location = 3 |
accommodation_2 | utility for school accommodation = 2 |
accommodation_3 | utility for school accommodation = 3 |
cgtraining_2 | utility for caregiver behavior training = 2 |
cgtraining_3 | utility for caregiver behavior training = 3 |
prvcomm_2 | utility for provider communication = 2 |
prvcomm_3 | utility for provider communication = 3 |
prvspec_2 | utility for provider specialty = 2 |
prvspec_3 | utility for provider specialty = 3 |
cost | utility for a $100 increase in monthly out-of-pocket costs |
09_population_parameters.csv
true values of the population parameters
This file contains the true values of the population parameters, obtained by calculating the mean and standard deviation of each of the 13 utility variables in the simulated population data. There are 13 rows for the means, followed by 13 rows for the standard deviations.
Column | Description |
---|---|
parameter | name of the parameter |
truth | true calculated value of the parameter |
12_choice_data.csv
choice data from the study sample
We selected a simple random sample of 500 individuals from the simulated population. We used each individual's simulated utility values to determine what choices the individual would make given the 18 choice tasks. For each individual and each choice task (500 x 18 = 9,000 rows), this file shows the attribute levels for the two alternatives in the choice task, as well as the choice that the individual "made."
Column | Description |
---|---|
id | study ID of the individual (1 to 500) |
case | unique row ID (1 to 9,000) |
task | index of the choice task (1 to 18) |
meds_2.1 | =1 if medication administration = 2 for alternative 1, 0 otherwise |
meds_3.1 | =1 if medication administration = 3 for alternative 1, 0 otherwise |
therapy_2.1 | =1 if therapy location = 2 for alternative 1, 0 otherwise |
therapy_3.1 | =1 if therapy location = 3 for alternative 1, 0 otherwise |
accommodation_2.1 | =1 if school accommodation = 2 for alternative 1, 0 otherwise |
accommodation_3.1 | =1 if school accommodation = 3 for alternative 1, 0 otherwise |
cgtraining_2.1 | =1 if caregiver behavior training = 2 for alternative 1, 0 otherwise |
cgtraining_3.1 | =1 if caregiver behavior training = 3 for alternative 1, 0 otherwise |
prvcomm_2.1 | =1 if provider communication = 2 for alternative 1, 0 otherwise |
prvcomm_3.1 | =1 if provider communication = 3 for alternative 1, 0 otherwise |
prvspec_2.1 | =1 if provider specialty = 2 for alternative 1, 0 otherwise |
prvspec_3.1 | =1 if provider specialty = 3 for alternative 1, 0 otherwise |
cost.1 | monthly out-of-pocket costs in hundreds of US dollars for alternative 1 |
meds_2.2 | =1 if medication administration = 2 for alternative 2, 0 otherwise |
meds_3.2 | =1 if medication administration = 3 for alternative 2, 0 otherwise |
therapy_2.2 | =1 if therapy location = 2 for alternative 2, 0 otherwise |
therapy_3.2 | =1 if therapy location = 3 for alternative 2, 0 otherwise |
accommodation_2.2 | =1 if school accommodation = 2 for alternative 2, 0 otherwise |
accommodation_3.2 | =1 if school accommodation = 3 for alternative 2, 0 otherwise |
cgtraining_2.2 | =1 if caregiver behavior training = 2 for alternative 2, 0 otherwise |
cgtraining_3.2 | =1 if caregiver behavior training = 3 for alternative 2, 0 otherwise |
prvcomm_2.2 | =1 if provider communication = 2 for alternative 2, 0 otherwise |
prvcomm_3.2 | =1 if provider communication = 3 for alternative 2, 0 otherwise |
prvspec_2.2 | =1 if provider specialty = 2 for alternative 2, 0 otherwise |
prvspec_3.2 | =1 if provider specialty = 3 for alternative 2, 0 otherwise |
cost.2 | monthly out-of-pocket costs in hundreds of US dollars for alternative 1 |
choice | the number of the alternative that the individual chose (1 or 2) |
14_results_cl.csv
multinomial logit (conditional logit) results
The multinomial logit (MNL) model estimates only the mean utility because it assumes constant utility across individuals with random error. This file contains the estimated mean utilities from the MNL model.
Column | Description |
---|---|
parameter | name of the parameter |
estimate | estimated mean |
stderr | standard error of the estimated mean |
15_results_rpl.csv
random parameter logit results
The random parameter logit (RPL) model estimates mean utilities and deviations around the means. In this case, the deviations are standard deviations because we simulated, and assumed in the RPL model, a normal distribution for each utility parameter. This file contains the estimated mean utilities and estimated standard deviations from the RPL model.
Column | Description |
---|---|
parameter | name of the parameter |
estimate | estimated mean or standard deviation |
stderr | standard error of the estimate |
17_sample_size_rpl_results.csv
sample size simulation results
We ran simulations to estimate the sample size required to estimate the parameters in the RPL model with 80% statistical power and an alpha level of .05. This file contains the simulation results.
Column | Description |
---|---|
n | sample size (250 to 2,500) |
sim | simulation number (1 to 50 for each sample size) |
parameter | name of the parameter |
estimate | estimated mean or standard deviation |
stderr | standard error of the estimate |
23_wtp_results_cl.csv
willingness-to-pay estimates for multinomial logit (conditional logit) model
This file contains estimates of participants' willingness to pay (WTP) for differences in attributes (for example, how much they would be willing to pay for their child to see one type of provider versus another). The estimates in this file came from the MNL model. We used 200 bootstrap replications to estimate 95% confidence intervals for the WTP estimates. Each row in the file represents one of the 200 bootstrap replications. Each column represents an attribute level (2 or 3). Each data cell contains a WTP estimate for a given attribute level, relative to the reference level (1) for that attribute.
Column | Description |
---|---|
meds_2 | WTP for medication administration = 2 |
meds_3 | WTP for medication administration = 3 |
therapy_2 | WTP for therapy location = 2 |
therapy_3 | WTP for therapy location = 3 |
accommodation_2 | WTP for school accommodation = 2 |
accommodation_3 | WTP for school accommodation = 3 |
cgtraining_2 | WTP for caregiver behavior training = 2 |
cgtraining_3 | WTP for caregiver behavior training = 3 |
prvcomm_2 | WTP for provider communication = 2 |
prvcomm_3 | WTP for provider communication = 3 |
prvspec_2 | WTP for provider specialty = 2 |
prvspec_3 | WTP for provider specialty = 3 |
24_wtp_results_rpl.csv
willingness-to-pay estimates for random parameter logit model
This file contains estimates of participants' willingness to pay (WTP) for differences in attributes (for example, how much they would be willing to pay for their child to see one type of provider versus another). The estimates in this file came from the RPL model. We used 200 bootstrap replications to estimate 95% confidence intervals for the WTP estimates. Each row in the file represents one of the 200 bootstrap replications. Each column represents an attribute level (2 or 3). Each data cell contains a WTP estimate for a given attribute level, relative to the reference level (1) for that attribute.
Column | Description |
---|---|
meds_2 | WTP for medication administration = 2 |
meds_3 | WTP for medication administration = 3 |
therapy_2 | WTP for therapy location = 2 |
therapy_3 | WTP for therapy location = 3 |
accommodation_2 | WTP for school accommodation = 2 |
accommodation_3 | WTP for school accommodation = 3 |
cgtraining_2 | WTP for caregiver behavior training = 2 |
cgtraining_3 | WTP for caregiver behavior training = 3 |
prvcomm_2 | WTP for provider communication = 2 |
prvcomm_3 | WTP for provider communication = 3 |
prvspec_2 | WTP for provider specialty = 2 |
prvspec_3 | WTP for provider specialty = 3 |
Description of Excel file contents
This section describes the contents of each individual Excel file, in the order in which the files are listed above.
These files include formulas, so they are considered software and hosted on Zenodo.
06_population_simulation_input.xlsx
parameters used for population simulation
This file contains 3 rows for each of the 7 attributes. Each set of 3 rows corresponds to the 3 levels used for an attribute in the original study by dosReis et al. (and shown in the online supplement of our article). The attribute names and descriptions follow:
- meds = medication administration
- therapy = therapy location
- accomodation = school accommodation
- cgtraining = caregiver behavior training
- prvcomm = provider communication
- prvspec = provider specialty
- cost = monthly out-of-pocket cost
The levels of each attribute are numbered 1 through 3.
Column | Description |
---|---|
attribute (dos Reis et al., 2017) | name and level of the attribute |
estimate (dosReis et al. 2017) | estimated utility of the attribute as reported by dosReis et al. |
SE (dosReis et al. 2017) | standard error of the estimated utility as reported by dosReis et al. |
means, converted to dummy coding | estimated mean utility for levels 2 and 3 of each categorical parameter (relative to level 1) and for the cost parameter |
SD, using abs(mean)/2 | standard deviation of utility, set at half the absolute value of the mean |
For each attribute, to estimate the mean utility for levels 2 and 3, we subtracted the level 1 parameter estimate from the level 2 and 3 parameter estimates, respectively.
To estimate the mean utility for each $100 increase in cost in the same way, we calculated the difference in utility between levels 1 and 2 ($50 and $150, respectively), calculated a third of the difference in utility between levels 2 and 3 ($150 and $450, respectively), and took the average.
We used the means and standard deviations in this file to simulate the population data from which we selected our sample.
19_attribute_importance_analysis.xlsx
Excel file to analyze attribute importance based on attribute level range and parameter estimates
We used this file to assess the relative importance of each attribute. We measured importance using the utility range of each attribute. The utility range is the difference between the highest and lowest estimated utilities associated with levels of the attribute. Because the cost attribute had the largest utility range, we expressed the importance of the other attributes relative to the importance of the cost attribute.
This file contains two tabs, one labeled CL for the conditional logit (multinomial logit) model, and the other labeled RPL for the random parameter logit model.
Column | Description |
---|---|
parameter | name of the parameter |
estimate | estimated mean from the MNL (CL) or RPL model |
attribute | name of the attribute associated with the parameter |
min | minimum level of the attribute |
max | maximum level of the attribute |
rng | placeholder for calculating max-min |
U(min) | estimated utility of the attribute's minimum level |
U(max) | estimated utility of the attribute's maximum level |
range | absolute difference between U(min) and U(max) |
range/max | importance of the attribute relative to the most important attribute, expressed as a percentage |
27_predicted_uptake.xlsx
Excel file to predict uptake of specific alternatives
We designed three treatment packages (Alternatives A, B, and C) with different sets of attributes (described in the online supplement). We then used the MNL results to predict the uptake of each package, that is, what percentage of the population would choose each package, given (a) a choice among the three or (b) a choice between the package and the status quo. We used this file to make the predictions. We also estimated how much the cost of each package would have to change in order for that package to have the same uptake as each other package.
The top portion of the file contains the data used to predict uptake:
Column | Description |
---|---|
parameter | name of the parameter (attribute) |
CL estimate | utility estimate from the conditional logit (multinomial logit) model |
Alternative A | The level of the attribute under Alternative A, dummy coded |
Alternative B | The level of the attribute under Alternative B, dummy coded |
Alternative C | The level of the attribute under Alternative C, dummy coded |
The bottom portion of the file contains the uptake predictions:
Label in Column A | Label in Column B | Value in Column Labeled Alternative A, B, or C |
---|---|---|
Uptake | utility | estimated utility of the alternative |
Uptake | exp(utility) | exponentiated utility |
Uptake | pred. prob. of selecting (versus others) | predicted probability of selecting the alternative, versus the others |
Uptake | pred. prob. of selecting (versus status quo) | predicted probability of selecting the alternative, versus the status quo |
Estimated change in cost to balance uptake | utility, except for cost | utility of the alternative minus the utility for cost |
Estimated change in cost to balance uptake | cost that would make uptake equal to that of Alternative A | cost that would make the alternative's uptake the same as that of Alternative A |
Estimated change in cost to balance uptake | cost that would make uptake equal to that of Alternative B | cost that would make the alternative's uptake the same as that of Alternative B |
Estimated change in cost to balance uptake | cost that would make uptake equal to that of Alternative C | cost that would make the alternative's uptake the same as that of Alternative C |
Source of data-generating parameters
We adapted the data-generating parameters from the following sources:
- dosReis, S., Mychailyszyn, M. P., Myers, M., & Riley, A. W. (2007). Coming to terms with ADHD: How urban African-American families come to seek care for their children. Psychiatric Services, 58(5), 636-641. https://doi.org/10.1176/ps.2007.58.5.636
- dosReis, S., Ng, X., Frosch, E., Reeves, G., Cunningham, C., & Bridges, J. (2015). Using best-worst scaling to measure caregiver preferences for managing their child’s ADHD: A pilot study. The Patient, 8(5), 423–431. https://doi.org/10.1007/s40271-014-0098-4
Code/Software
Most of the code in this submission was implemented in R version 4.3.2. The single SAS program (which we used to identify an alternate experimental design) was implemented in SAS Version 9.4. For some analyses, we used Microsoft Excel Version 16.84.
Methods
In the worked example, we used simulated data to examine caregiver preferences for 7 treatment attributes (medication administration, therapy location, school accommodation, caregiver behavior training, provider communication, provider specialty, and monthly out-of-pocket costs) identified by dosReis and colleagues in a previous DCE. We employed an orthogonal design with 1 continuous variable (cost) and 12 dummy-coded variables (representing the levels of the remaining attributes, which were categorical). Using the parameter estimates published by dosReis et al., with slight adaptations, we simulated utility values for a population of 100,000 people, then selected a sample of 500 for analysis. Relying on random utility theory, we used the mlogit package in R to estimate the MNL and RPL models, using 5,000 Halton draws for simulated maximum likelihood estimation of the RPL model. In addition to estimating the utility parameters, we measured the relative importance of each attribute, estimated caregivers’ willingness to pay (WTP) for differences in attributes (e.g., how much they would be willing to pay for their child to see one type of provider versus another) with bootstrapped 95% confidence intervals, and predicted the uptake of three treatment packages with different sets of attributes. This submission includes both the simulated source data and the processed results. The online supplement of the primary article describes the methods in greater detail.