| Title: | Choice Models with Economic Foundation |
|---|---|
| Description: | Implements choice models based on economic theory, including estimation using Markov chain Monte Carlo (MCMC), prediction, and more. Its usability is inspired by ideas from 'tidyverse'. Models include versions of the Hierarchical Multinomial Logit and Multiple Discrete-Continous (Volumetric) models with and without screening. The foundations of these models are described in Allenby, Hardt and Rossi (2019) <doi:10.1016/bs.hem.2019.04.002>. Models with conjunctive screening are described in Kim, Hardt, Kim and Allenby (2022) <doi:10.1016/j.ijresmar.2022.04.001>. Models with set-size variation are described in Hardt and Kurz (2020) <doi:10.2139/ssrn.3418383>. |
| Authors: | Nino Hardt [aut, cre] |
| Maintainer: | Nino Hardt <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 0.2.4 |
| Built: | 2024-11-15 06:50:57 UTC |
| Source: | CRAN |
Get the attribute of an object
obj %.% attrnameobj %.% attrname
obj |
The object to get the attribute from. |
attrname |
The name of the attribute to get. |
The attribute of the object.
obj <- list(a = 1, b = 2) attributes(obj)$test="hello" `%.%`(obj, "test")obj <- list(a = 1, b = 2) attributes(obj)$test="hello" `%.%`(obj, "test")
Discrete Choice Predictions (HMNL)
dd_dem(dd, est, prob = FALSE, cores = NULL)dd_dem(dd, est, prob = FALSE, cores = NULL)
dd |
tibble with long-format choice data |
est |
estimation object |
prob |
logical, report probabilities instead of demand |
cores |
cores |
Draws of expected choice
dd_est_hmnl() to generate demand predictions based on this model
data(icecream_discrete) icecream_est <- icecream_discrete %>% filter(id<10) %>% dd_est_hmnl(R=4, cores=2) #demand prediction icecream_dempred <- icecream_discrete %>% filter(id<10) %>% dd_dem(icecream_est, cores=2)data(icecream_discrete) icecream_est <- icecream_discrete %>% filter(id<10) %>% dd_est_hmnl(R=4, cores=2) #demand prediction icecream_dempred <- icecream_discrete %>% filter(id<10) %>% dd_dem(icecream_est, cores=2)
Discrete Choice Predictions (HMNL with attribute-based screening)
dd_dem_sr(dd, est, prob = FALSE, cores = NULL)dd_dem_sr(dd, est, prob = FALSE, cores = NULL)
dd |
data |
est |
est |
prob |
logical, report probabilities instead of demand |
cores |
cores |
Draws of expected choice
dd_est_hmnl_screen() to generate demand predictions based on this model
data(icecream_discrete) icecream_est <- icecream_discrete %>% filter(id<20) %>% dd_est_hmnl_screen(R=10, cores=2) #demand prediction icecream_dempred <- icecream_discrete %>% filter(id<20) %>% dd_dem_sr(icecream_est, cores=2)data(icecream_discrete) icecream_est <- icecream_discrete %>% filter(id<20) %>% dd_est_hmnl_screen(R=10, cores=2) #demand prediction icecream_dempred <- icecream_discrete %>% filter(id<20) %>% dd_dem_sr(icecream_est, cores=2)
Estimate discrete choice model (HMNL)
dd_est_hmnl( dd, R = 1e+05, keep = 10, cores = NULL, control = list(include_data = TRUE) )dd_est_hmnl( dd, R = 1e+05, keep = 10, cores = NULL, control = list(include_data = TRUE) )
dd |
discrete choice data (long format) |
R |
draws |
keep |
thinning |
cores |
no of CPU cores to use (default: auto-detect) |
control |
list containing additional settings |
est ec-draw object (List)
dd_dem() to generate demand predictions based on this model
data(icecream_discrete) icecream_est <- icecream_discrete %>% dd_est_hmnl(R=20, cores=2)data(icecream_discrete) icecream_est <- icecream_discrete %>% dd_est_hmnl(R=20, cores=2)
Estimate discrete choice model (HMNL, attribute-based screening (not including price))
dd_est_hmnl_screen( dd, price_screen = TRUE, R = 1e+05, keep = 10, cores = NULL, control = list(include_data = TRUE) )dd_est_hmnl_screen( dd, price_screen = TRUE, R = 1e+05, keep = 10, cores = NULL, control = list(include_data = TRUE) )
dd |
discrete choice data (long format) |
price_screen |
A logical, indicating whether price tag screening should be estimated |
R |
draws |
keep |
thinning |
cores |
no of CPU cores to use (default: auto-detect) |
control |
list containing additional settings |
est ec-draw object (List)
dd_dem_sr() to generate demand predictions based on this model
data(icecream_discrete) icecream_est <- icecream_discrete %>% dplyr::filter(id<20) %>% dd_est_hmnl_screen(R=20, cores=2)data(icecream_discrete) icecream_est <- icecream_discrete %>% dplyr::filter(id<20) %>% dd_est_hmnl_screen(R=20, cores=2)
Log-Likelihood for compensatory hmnl model
dd_LL(draw, dd, fromdraw = 1)dd_LL(draw, dd, fromdraw = 1)
draw |
A list, 'echoice2' draws object |
dd |
A tibble, tidy choice data (before dummy-coding) |
fromdraw |
An integer, from which draw onwards to compute LL (i.e., excl. burnin) |
N x Draws Matrix of log-Likelihood values
data(icecream_discrete) #fit model icecream_est <- icecream_discrete %>% dd_est_hmnl(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-dd_LL(icecream_est, icecream_discrete, fromdraw = 2)data(icecream_discrete) #fit model icecream_est <- icecream_discrete %>% dd_est_hmnl(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-dd_LL(icecream_est, icecream_discrete, fromdraw = 2)
Log-Likelihood for screening hmnl model
dd_LL_sr(draw, dd, fromdraw = 1)dd_LL_sr(draw, dd, fromdraw = 1)
draw |
A list, 'echoice2' draws object |
dd |
A tibble, tidy choice data (before dummy-coding) |
fromdraw |
An integer, from which draw onwards to compute LL (i.e., excl. burnin) |
N x Draws Matrix of log-Likelihood values
data(icecream_discrete) #fit model icecream_est <- icecream_discrete %>% dd_est_hmnl_screen(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-dd_LL_sr(icecream_est, icecream_discrete, fromdraw = 2)data(icecream_discrete) #fit model icecream_est <- icecream_discrete %>% dd_est_hmnl_screen(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-dd_LL_sr(icecream_est, icecream_discrete, fromdraw = 2)
Create dummy variables within a tibble
dummify(dat, sel)dummify(dat, sel)
dat |
A |
sel |
A character vector with the name(s) of the variables to be dummied. |
tibble with dummy variables
mytest=data.frame(A=factor(c('a','a','b','c','c')), B=1:5) dummify(mytest,"A")mytest=data.frame(A=factor(c('a','a','b','c','c')), B=1:5) dummify(mytest,"A")
Dummy-code a categorical variable
dummyvar(data)dummyvar(data)
data |
one column of categorical data to be dummy-coded |
tibble with dummy variables
mytest=data.frame(attribute=factor(c('a','a','b','c','c'))) dummyvar(mytest)mytest=data.frame(attribute=factor(c('a','a','b','c','c'))) dummyvar(mytest)
Generate MU_theta boxplot
ec_boxplot_MU(draws, burnin = 100)ec_boxplot_MU(draws, burnin = 100)
draws |
A list, 'echoice2' draws object |
burnin |
burn-in to remove |
A ggplot2 plot containing traceplots of draws
ec_trace_MU() to obtain traceplot
## Not run: data(icecream) #run MCMC sampler (use way more than 50 draws for actual use icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=20, cores=2) ec_boxplot_MU(icecream_est, burnin=1) ## End(Not run)## Not run: data(icecream) #run MCMC sampler (use way more than 50 draws for actual use icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=20, cores=2) ec_boxplot_MU(icecream_est, burnin=1) ## End(Not run)
Generate Screening probability boxplot
ec_boxplot_screen(draws, burnin = 100)ec_boxplot_screen(draws, burnin = 100)
draws |
A list, 'echoice2' draws object, from a model with attribute-based screening |
burnin |
burn-in to remove |
A ggplot2 plot containing traceplots of draws
ec_draws_MU() to obtain MU_theta draws,
ec_trace_screen() to generate traceplot
data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_scr_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm_screen(R=20, cores=2) ec_boxplot_screen(icecream_scr_est, burnin = 1)data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_scr_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm_screen(R=20, cores=2) ec_boxplot_screen(icecream_scr_est, burnin = 1)
Aggregate demand draws, e.g. from individual-choice occasion-alternative level to individual level. (using the new demand draw format)
ec_dem_aggregate(de,groupby)ec_dem_aggregate(de,groupby)
de |
demand draws |
groupby |
groupby grouping variables (as (vector of) string(s)) |
Aggregated demand predictions
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=4, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #aggregate brand_lvl_pred_demand <- icecream_predicted_demand %>% ec_dem_aggregate("Brand")data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=4, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #aggregate brand_lvl_pred_demand <- icecream_predicted_demand %>% ec_dem_aggregate("Brand")
This function obtains proper posterior fit statistics. It computes the difference between true demand and each draw from the demand posterior. Then, fit statistics are obtained.
ec_dem_eval(de)ec_dem_eval(de)
de |
demand draws (output from vd_dem_x function) |
Predictive fit statistics (MAE, MSE, RAE, bias, hit-probability)
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=20, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<100) %>% vd_dem_vdm(icecream_est) #evaluate in-sample fit (note: too few draws for good results) ec_dem_eval(icecream_predicted_demand)
Adds summaries of posterior draws of demand to tibble. (using the new demand draw format)
ec_dem_summarise(de,quantiles) ec_dem_summarize(de, quantiles = c(0.05, 0.95))ec_dem_summarise(de,quantiles) ec_dem_summarize(de, quantiles = c(0.05, 0.95))
de |
demand draws |
quantiles |
Quantiles for Credibility Intervals (default: 90% interval) |
Summary of demand predictions
data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #aggregate brand_lvl_pred_demand <- icecream_predicted_demand %>% ec_dem_aggregate("Brand") #summarise brand_lvl_pred_demand %>% ec_dem_summarise()data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #aggregate brand_lvl_pred_demand <- icecream_predicted_demand %>% ec_dem_aggregate("Brand") #summarise brand_lvl_pred_demand %>% ec_dem_summarise()
This helper function creates demand curves
ec_demcurve( ec_long, focal_product, rel_pricerange, dem_fun, draws, epsilon_not = NULL )ec_demcurve( ec_long, focal_product, rel_pricerange, dem_fun, draws, epsilon_not = NULL )
ec_long |
choice scenario (discrete or volumetric) |
focal_product |
Logical vector picking the focal product for which to create a demand curve |
rel_pricerange |
Price range, relative to base case price; this is used to create demand curve |
dem_fun |
demand function (e.g., |
draws |
ec-draws object (e.g., output from |
epsilon_not |
(optional) error realisatins (this helps make curves look smother for voumetric models) |
List containing aggregate demand quantities for each scenario defined by rel_pricerange
ec_gen_err_normal() to generate error realization from Normal distribution,
ec_gen_err_ev1() to generate error realization from EV1 distribution
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=20, keep=1, cores=2) #demand at different price points dem_scenarios<- ec_demcurve(icecream%>% dplyr::filter(id<100), icecream%>% dplyr::filter(id<100) %>% pull('Brand')=="Store", c(.75,1,1.25),vd_dem_vdm,icecream_est) #optional plot # dem_scenarios %>% # do.call('rbind',.) %>% # ggplot(aes(x=scenario,y=`E(demand)`,color=Flavor)) + geom_line()data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=20, keep=1, cores=2) #demand at different price points dem_scenarios<- ec_demcurve(icecream%>% dplyr::filter(id<100), icecream%>% dplyr::filter(id<100) %>% pull('Brand')=="Store", c(.75,1,1.25),vd_dem_vdm,icecream_est) #optional plot # dem_scenarios %>% # do.call('rbind',.) %>% # ggplot(aes(x=scenario,y=`E(demand)`,color=Flavor)) + geom_line()
This helper function creates demand curves
ec_demcurve_cond_dem( ec_long, focal_product, rel_pricerange, dem_fun, draws, epsilon_not = NULL )ec_demcurve_cond_dem( ec_long, focal_product, rel_pricerange, dem_fun, draws, epsilon_not = NULL )
ec_long |
choice scenario (discrete or volumetric) |
focal_product |
Logical vector picking the focal product for which to create a demand curve |
rel_pricerange |
Price range, relative to base case price; this is used to create demand curve |
dem_fun |
demand function (e.g., |
draws |
ec-draws object (e.g., output from |
epsilon_not |
(optional) error realisatins (this helps make curves look smother for voumetric models) |
List containing aggregate demand quantities for each scenario defined by rel_pricerange
ec_gen_err_normal() to generate error realization from Normal distribution,
ec_gen_err_ev1() to generate error realization from EV1 distribution
data(icecream) #run MCMC sampler (use way more draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=2, keep=1, cores=2) #demand at different price points conddem_scenarios<- ec_demcurve_cond_dem(icecream%>% dplyr::filter(id<20), icecream%>% dplyr::filter(id<20) %>% pull('Brand')=="Store", c(.75,1),vd_dem_vdm,icecream_est)data(icecream) #run MCMC sampler (use way more draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=2, keep=1, cores=2) #demand at different price points conddem_scenarios<- ec_demcurve_cond_dem(icecream%>% dplyr::filter(id<20), icecream%>% dplyr::filter(id<20) %>% pull('Brand')=="Store", c(.75,1),vd_dem_vdm,icecream_est)
This helper function creates demand curves
ec_demcurve_inci( ec_long, focal_product, rel_pricerange, dem_fun, draws, epsilon_not = NULL )ec_demcurve_inci( ec_long, focal_product, rel_pricerange, dem_fun, draws, epsilon_not = NULL )
ec_long |
choice scenario (discrete or volumetric) |
focal_product |
Logical vector picking the focal product for which to create a demand curve |
rel_pricerange |
Price range, relative to base case price; this is used to create demand curve |
dem_fun |
demand function (e.g., |
draws |
ec-draws object (e.g., output from |
epsilon_not |
(optional) error realisatins (this helps make curves look smother for voumetric models) |
List containing aggregate demand quantities for each scenario defined by rel_pricerange
ec_gen_err_normal() to generate error realization from Normal distribution,
ec_gen_err_ev1() to generate error realization from EV1 distribution
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=20, keep=1, cores=2) #demand at different price points inci_scenarios<- ec_demcurve_inci(icecream%>% dplyr::filter(id<50), icecream%>% dplyr::filter(id<50) %>% pull('Brand')=="Store", c(.75,1,1.25),vd_dem_vdm,icecream_est)data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=20, keep=1, cores=2) #demand at different price points inci_scenarios<- ec_demcurve_inci(icecream%>% dplyr::filter(id<50), icecream%>% dplyr::filter(id<50) %>% pull('Brand')=="Store", c(.75,1,1.25),vd_dem_vdm,icecream_est)
Obtain MU_theta draws
ec_draws_MU(draws)ec_draws_MU(draws)
draws |
A list, 'echoice2' draws object |
A tibble, long format, draws of MU
ec_draws_screen() to obtain screening parameter draws (where applicable),
ec_trace_MU() to generate a traceplot of MU_theta draws
data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=20, cores=2) ec_draws_MU(icecream_est)data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=20, cores=2) ec_draws_MU(icecream_est)
Obtain Screening probability draws
ec_draws_screen(draws)ec_draws_screen(draws)
draws |
A list, 'echoice2' draws object |
A tibble, long format, draws of MU
ec_draws_MU() to obtain MU_theta draws,
ec_trace_screen() to generate a traceplot of screening draws
data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_scr_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm_screen(R=20, cores=2) ec_draws_screen(icecream_scr_est)data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_scr_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm_screen(R=20, cores=2) ec_draws_screen(icecream_scr_est)
Obtain upper level model estimates
ec_estimates_MU(est, quantiles = c(0.05, 0.95))ec_estimates_MU(est, quantiles = c(0.05, 0.95))
est |
is an 'echoice2' draw object (list) |
quantiles |
quantile for CI |
tibble with MU (upper level) summaries
data(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=20, cores=2) #Upper-level summary icecream_est %>% ec_estimates_MUdata(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=20, cores=2) #Upper-level summary icecream_est %>% ec_estimates_MU
Summarize attribute-based screening parameters from an attribute-based screening model in 'echoice2'
ec_estimates_screen(est, quantiles = c(0.05, 0.95))ec_estimates_screen(est, quantiles = c(0.05, 0.95))
est |
is an 'echoice2' draw object (list) from a model with attribute-based screening |
quantiles |
quantile for CI |
tibble with screening summaries
#run MCMC sampler (use way more than 20 draws for actual use) data(icecream) est_scr_icecream <- vd_est_vdm_screen(icecream%>%dplyr::filter(id<30), R=20, cores=2) #summarise draws of screening probabilities ec_estimates_screen(est_scr_icecream) #Note: There is no variance in this illustrative example - more draws are needed#run MCMC sampler (use way more than 20 draws for actual use) data(icecream) est_scr_icecream <- vd_est_vdm_screen(icecream%>%dplyr::filter(id<30), R=20, cores=2) #summarise draws of screening probabilities ec_estimates_screen(est_scr_icecream) #Note: There is no variance in this illustrative example - more draws are needed
Obtain posterior mean estimates of upper level covariance
ec_estimates_SIGMA(est)ec_estimates_SIGMA(est)
est |
is an 'echoice2' draw object (list) |
estimates of upper level covariance
data(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=20, cores=2) icecream_est %>% ec_estimates_SIGMA %>% round(2)data(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=20, cores=2) icecream_est %>% ec_estimates_SIGMA %>% round(2)
Obtain posterior mean estimates of upper level correlations
ec_estimates_SIGMA_corr(est)ec_estimates_SIGMA_corr(est)
est |
is an 'echoice2' draw object (list) |
estimates of upper level correlations
data(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=20, cores=2) icecream_est %>% ec_estimates_SIGMA_corr %>% round(2)data(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=20, cores=2) icecream_est %>% ec_estimates_SIGMA_corr %>% round(2)
Simulate error realization from EV1 distribution
ec_gen_err_ev1(ec_dem, draws, seed = NULL)ec_gen_err_ev1(ec_dem, draws, seed = NULL)
ec_dem |
discrete or volumetric choice data, with or without x |
draws |
draws from volumetric demand model |
seed |
seed for reproducible error realisations; seet is automatically reset of running this function |
error realizations
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=100, keep=1, cores=2) #generate error realizations errs<- ec_gen_err_ev1(icecream %>% dplyr::filter(id<100), icecream_est, seed=123)data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=100, keep=1, cores=2) #generate error realizations errs<- ec_gen_err_ev1(icecream %>% dplyr::filter(id<100), icecream_est, seed=123)
Simulate error realization from Normal distribution
ec_gen_err_normal(ec_dem, draws, seed = NULL)ec_gen_err_normal(ec_dem, draws, seed = NULL)
ec_dem |
discrete or volumetric choice data, with or without x |
draws |
draws from volumetric demand model |
seed |
seed for reproducible error realisations; seet is automatically reset of running this function |
error realizations
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, keep=1, error_dist = "Normal", cores=2) #generate error realizations errs<- ec_gen_err_normal(icecream %>% dplyr::filter(id<10), icecream_est, seed=123)data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, keep=1, error_dist = "Normal", cores=2) #generate error realizations errs<- ec_gen_err_normal(icecream %>% dplyr::filter(id<10), icecream_est, seed=123)
This is a helper function to quickly obtain log marginal density from a draw object
ec_lmd_NR(est)ec_lmd_NR(est)
est |
'echoice2' draw object |
Draws are split in 4 equal parts from start to finish, and LMD is computed for each part. This helps to double-check convergence.
tibble with LMDs (first 25% of draws, next 25% of draws, ...)
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=20, cores=2) #obtain LMD by quartile of draws ec_lmd_NR(icecream_est)data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=20, cores=2) #obtain LMD by quartile of draws ec_lmd_NR(icecream_est)
Convert "list of lists" format to long "tidy" format
ec_lol_tidy1(data_lol, X = "X", y = "y")ec_lol_tidy1(data_lol, X = "X", y = "y")
data_lol |
A list of data frames containing design matrices and response vectors |
X |
The column name of the design matrix, default: "X" |
y |
The column name of the response vector, default: "y" |
A tidy data frame with columns for each design matrix column, the response vector, and an id column indicating which data frame the row came from
loldata<-list() loldata[[1]]=list() loldata[[1]]$y = c(1,2) loldata[[1]]$X= data.frame(brand1=c(1,0, 1,0),brand2=c(0,1, 0,1),price=c(1,2)) loldata[[2]]=list() loldata[[2]]$y = c(1,1) loldata[[2]]$X= data.frame(brand1=c(1,0, 1,0),brand2=c(0,1, 0,1),price=c(1,2)) ec_lol_tidy1(loldata)loldata<-list() loldata[[1]]=list() loldata[[1]]$y = c(1,2) loldata[[1]]$X= data.frame(brand1=c(1,0, 1,0),brand2=c(0,1, 0,1),price=c(1,2)) loldata[[2]]=list() loldata[[2]]$y = c(1,1) loldata[[2]]$X= data.frame(brand1=c(1,0, 1,0),brand2=c(0,1, 0,1),price=c(1,2)) ec_lol_tidy1(loldata)
Adds summaries of posterior draws of demand to tibble. (using the new demand draw format)
ec_screen_summarise(sc, quantiles = c(0.05, 0.95)) ec_screen_summarize(sc, quantiles = c(0.05, 0.95))ec_screen_summarise(sc, quantiles = c(0.05, 0.95)) ec_screen_summarize(sc, quantiles = c(0.05, 0.95))
sc |
tibble containing screening draws in .screendraws |
quantiles |
Quantiles for Credibility Intervals (default: 90% interval) |
Summary of screening probabilities
data(icecream) icecream_est <- icecream %>% vd_est_vdm_screen(R=20, price_screen=TRUE, cores=2) #consideration set by respondent cons_ss <- ec_screenprob_sr(icecream, icecream_est, cores=2) %>% group_by(id, task) %>% summarise(.screendraws=list(purrr::reduce(.screendraws ,`+`))) %>% ec_screen_summarise() %>% group_by(id) %>% summarise(n_screen=mean(`E(screening)`))data(icecream) icecream_est <- icecream %>% vd_est_vdm_screen(R=20, price_screen=TRUE, cores=2) #consideration set by respondent cons_ss <- ec_screenprob_sr(icecream, icecream_est, cores=2) %>% group_by(id, task) %>% summarise(.screendraws=list(purrr::reduce(.screendraws ,`+`))) %>% ec_screen_summarise() %>% group_by(id) %>% summarise(n_screen=mean(`E(screening)`))
Obtain draws of screening probabilities of choiec alternatives
ec_screenprob_sr(xd, est, cores=NULL)ec_screenprob_sr(xd, est, cores=NULL)
xd |
data |
est |
ec-model draws |
cores |
(optional) cores |
Draws of screening probabilities of choice alternatives
data(icecream) icecream_est <- icecream %>% filter(id<10) %>% vd_est_vdm_screen(R=10, price_screen=TRUE, cores=2) ec_screenprob_sr(icecream %>% filter(id<10), icecream_est, cores=2)data(icecream) icecream_est <- icecream %>% filter(id<10) %>% vd_est_vdm_screen(R=10, price_screen=TRUE, cores=2) ec_screenprob_sr(icecream %>% filter(id<10), icecream_est, cores=2)
Summarize attributes and levels in tidy choice data containing categorical attributes (before dummy-coding)
ec_summarize_attrlvls(data_in) ec_summarise_attrlvls(data_in)ec_summarize_attrlvls(data_in) ec_summarise_attrlvls(data_in)
data_in |
A tibble, containing long-format choice data |
This functions looks for categorical attributes and summaries their levels This is helpful when evaluating a new choice data file.
A tibble with one row per attribute, and a list of the levels
data(icecream) ec_summarize_attrlvls(icecream)data(icecream) ec_summarize_attrlvls(icecream)
Generate MU_theta traceplot
ec_trace_MU(draws, burnin = 100)ec_trace_MU(draws, burnin = 100)
draws |
A list, 'echoice2' draws object |
burnin |
burn-in to remove |
A ggplot2 plot containing traceplots of draws
ec_boxplot_MU() to obtain boxplot
## Not run: data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, cores=2) ec_trace_MU(icecream_est) ## End(Not run)## Not run: data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, cores=2) ec_trace_MU(icecream_est) ## End(Not run)
Generate Screening probability traceplots
ec_trace_screen(draws, burnin = 100)ec_trace_screen(draws, burnin = 100)
draws |
A list, 'echoice2' draws object, from a model with attribute-based screening |
burnin |
burn-in to remove |
A ggplot2 plot containing traceplots of draws
ec_draws_MU() to obtain MU_theta draws,
ec_boxplot_screen() to generate boxplot
## Not run: data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_scr_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm_screen(R=20, cores=2) ec_trace_screen(icecream_scr_est, burnin=1) ## End(Not run)## Not run: data(icecream) #run MCMC sampler (use way more than 20 draws for actual use icecream_scr_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm_screen(R=20, cores=2) ec_trace_screen(icecream_scr_est, burnin=1) ## End(Not run)
Given a set of dummy variables, this function converts them into a single categorical variable. The categorical variable is created by determining which variables are active (i.e. have a value of 1) for each observation and assigning a category based on the set of active variables. If necessary, a reference level can be specified to ensure that all possible categories are represented. Often, all brands of a brand attribute are added as brand intercepts, while other categorical attributes are coded with respect to a reference level.
ec_undummy(data_in, set_members, attribute_name, ref_level = NULL)ec_undummy(data_in, set_members, attribute_name, ref_level = NULL)
data_in |
a data frame containing the dummy variables |
set_members |
a character vector of the names of the dummy variables |
attribute_name |
a character string representing the name of the new categorical variable |
ref_level |
a character string representing the name of the reference level. If specified, a new dummy variable will be created for this level, and it will be used as the reference category for the categorical variable. Defaults to NULL. |
a data frame with the same columns as data_in, except for the
dummy variables in set_members, which are replaced with the new
categorical variable attribute_name
minidata=structure(list(id = c("1", "1", "1", "1", "2", "2", "2", "2"), task = c(1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L), alt = c(1L, 2L, 1L, 2L, 1L, 2L, 1L, 2L), brand1 = c(1, 0, 1, 0, 1, 0, 1, 0), brand2 = c(0, 1, 0, 1, 0, 1, 0, 1), price = c(1, 2, 1, 2, 1, 2, 1, 2), x = c(1, 0, 0, 1, 1, 0, 1, 0)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA, -8L)) minidata %>% ec_undummy(c('brand1','brand2'),"brand")minidata=structure(list(id = c("1", "1", "1", "1", "2", "2", "2", "2"), task = c(1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L), alt = c(1L, 2L, 1L, 2L, 1L, 2L, 1L, 2L), brand1 = c(1, 0, 1, 0, 1, 0, 1, 0), brand2 = c(0, 1, 0, 1, 0, 1, 0, 1), price = c(1, 2, 1, 2, 1, 2, 1, 2), x = c(1, 0, 0, 1, 1, 0, 1, 0)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA, -8L)) minidata %>% ec_undummy(c('brand1','brand2'),"brand")
Convert dummy-coded variables to low/high factor
ec_undummy_lowhigh(vec_in)ec_undummy_lowhigh(vec_in)
vec_in |
A vector of dummy-coded variables (0/1) |
A factor vector with levels "low" and "high"
ec_undummy_lowhigh(c(0,1,0,1,1))ec_undummy_lowhigh(c(0,1,0,1,1))
Convert dummy-coded variables to low/medium/high factor
ec_undummy_lowmediumhigh(vec_in)ec_undummy_lowmediumhigh(vec_in)
vec_in |
A vector of dummy-coded variables (0/1/2) |
A factor vector with levels "low", "medium" and "high"
ec_undummy_lowmediumhigh(c(0,1,2,1,0,2))ec_undummy_lowmediumhigh(c(0,1,2,1,0,2))
Convert dummy-coded variables to yes/no factor
ec_undummy_yesno(vec_in)ec_undummy_yesno(vec_in)
vec_in |
A vector of dummy-coded variables (0/1) |
A factor vector with levels "no" and "yes"
ec_undummy_yesno(c(0,1,0,1,1))ec_undummy_yesno(c(0,1,0,1,1))
Converts a vector of choices into a long format data frame, where each row represents a single choice and contains the choice status for each alternative.
ec_util_choice_to_long(myvec, all_index)ec_util_choice_to_long(myvec, all_index)
myvec |
A vector of choices, where each element represents the index of the chosen alternative. |
all_index |
A vector of all the possible alternative indices. |
A tibble with columns 'x', 'task', and 'alt', where 'x' is a binary indicator of whether the alternative was chosen or not, 'task' is the task index, and 'alt' is the alternative index.
#There are 3 alternatives in this task. #Since there are 3 observations in myvec, there are 3 tasks total. ec_util_choice_to_long(c(1, 2, 1), c(1, 2, 3))#There are 3 alternatives in this task. #Since there are 3 observations in myvec, there are 3 tasks total. ec_util_choice_to_long(c(1, 2, 1), c(1, 2, 3))
This function finds pairs of columns in a data frame that are mutually exclusive, i.e., that never have positive values at the same time.
ec_util_dummy_mutualeclusive(data_in, filtered = TRUE)ec_util_dummy_mutualeclusive(data_in, filtered = TRUE)
data_in |
A data frame containing the data. |
filtered |
A logical value indicating whether to return only the mutually exclusive pairs (TRUE) or all pairs (FALSE). Default is TRUE. |
A tibble containing all pairs of mutually exclusive columns in the data frame.
minidata=structure(list(id = c("1", "1", "1", "1", "2", "2", "2", "2"), task = c(1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L), alt = c(1L, 2L, 1L, 2L, 1L, 2L, 1L, 2L), brand1 = c(1, 0, 1, 0, 1, 0, 1, 0), brand2 = c(0, 1, 0, 1, 0, 1, 0, 1), price = c(1, 2, 1, 2, 1, 2, 1, 2), x = c(1, 0, 0, 1, 1, 0, 1, 0)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA, -8L)) ec_util_dummy_mutualeclusive(minidata)minidata=structure(list(id = c("1", "1", "1", "1", "2", "2", "2", "2"), task = c(1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L), alt = c(1L, 2L, 1L, 2L, 1L, 2L, 1L, 2L), brand1 = c(1, 0, 1, 0, 1, 0, 1, 0), brand2 = c(0, 1, 0, 1, 0, 1, 0, 1), price = c(1, 2, 1, 2, 1, 2, 1, 2), x = c(1, 0, 0, 1, 1, 0, 1, 0)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA, -8L)) ec_util_dummy_mutualeclusive(minidata)
Obtain attributes and levels from tidy choice data with dummies
get_attr_lvl(tdc)get_attr_lvl(tdc)
tdc |
A tibble with choice data |
tibble
mytest=data.frame(A=factor(c('a','a','b','c','c')), B=1:5) dummied_data = dummify(mytest,"A") get_attr_lvl(dummied_data)mytest=data.frame(A=factor(c('a','a','b','c','c')), B=1:5) dummied_data = dummify(mytest,"A") get_attr_lvl(dummied_data)
Volumetric Conjoint data, ice cream category
Data from volumetric conjoint analysis in the ice cream category. 300 respondents total. Volumetric demand in units of 4 ounces each. Attributes include brand name, flavor, and container size.
Discrete-Choice Conjoint data, ice cream category
Data from discrete choice conjoint analysis in the ice cream category. 300 respondents total. Attributes include brand name, flavor, and container size.
This function uses the quick-and-dirty Newton-Raftery approximation for log-marginal-density.
logMargDenNRu(ll)logMargDenNRu(ll)
ll |
A vector of log-likelihood values (i.e., draws) |
Approximation of LMD based on Newton-Raftery. It is not the most accurate, but a very fast method.
A single numeric value representing the log marginal density
logll_values <- c(-4000, -4001, -4002) logMargDenNRu(logll_values)logll_values <- c(-4000, -4001, -4002) logMargDenNRu(logll_values)
Volumetric Conjoint data, pizza category
Data from volumetric conjoint analysis in the frozen pizza category.
Makes sure the factor levels in data_new are aligned with data_old
This is helpful for demand simulations.
prep_newprediction(data_new, data_old)prep_newprediction(data_new, data_old)
data_new |
New long-format choice data |
data_old |
Old long-format choice data |
long-format choice data
data(icecream) prep_newprediction(icecream, icecream)data(icecream) prep_newprediction(icecream, icecream)
This adds a unique product identifier to demand draw objects.
vd_add_prodid(de)vd_add_prodid(de)
de |
demand draws |
est
data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=4, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #add prodid icecream_predicted_demand_w_id<-icecream_predicted_demand %>% vd_add_prodiddata(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=4, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #add prodid icecream_predicted_demand_w_id<-icecream_predicted_demand %>% vd_add_prodid
Adds summaries of posterior draws of demand to tibble. (using the new demand draw format)
vd_dem_summarise(de, quantiles = c(0.05, 0.95)) vd_dem_summarize(de, quantiles = c(0.05, 0.95))vd_dem_summarise(de, quantiles = c(0.05, 0.95)) vd_dem_summarize(de, quantiles = c(0.05, 0.95))
de |
demand draws |
quantiles |
Quantiles for Credibility Intervals (default: 90% interval) |
Summary of demand predictions
data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #aggregate brand_lvl_pred_demand <- icecream_predicted_demand %>% ec_dem_aggregate("Brand") #summarise brand_lvl_pred_demand %>% vd_dem_summarise()data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm(icecream_est) #aggregate brand_lvl_pred_demand <- icecream_predicted_demand %>% ec_dem_aggregate("Brand") #summarise brand_lvl_pred_demand %>% vd_dem_summarise()
Generating demand predictions for volumetric demand model.
Reminder: there is no closed-form solution for demand, thus we need to integrate not only over the posterior distribution of parameters and the error distribution.
The function outputs a tibble containing id, task, alt, p, attributes, draws from the posterior of demand.
Error realizations can be pre-supplied to the epsilon_not. This helps create smooth demand curves or conduct optimization.
vd_dem_vdm( vd, est, epsilon_not = NULL, error_dist = NULL, tidy = TRUE, cores = NULL )vd_dem_vdm( vd, est, epsilon_not = NULL, error_dist = NULL, tidy = TRUE, cores = NULL )
vd |
data |
est |
ec-model draws |
epsilon_not |
(optional) error realizations |
error_dist |
(optional) A string defining the error term distribution (default: 'EV1') |
tidy |
(optional) apply 'echoice2' tidier (default: TRUE) |
cores |
(optional) cores (default: auto-detect) |
Draws of expected demand
prep_newprediction() to match vd's factor levels,
ec_gen_err_ev1() for pre-generating error realizations and
vd_est_vdm() for estimating the corresponding model
data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<20) %>% vd_dem_vdm(icecream_est, cores=2) #column .demdraws contains draws from posterior of predicted demanddata(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<20) %>% vd_dem_vdm(icecream_est, cores=2) #column .demdraws contains draws from posterior of predicted demand
Generating demand predictions for volumetric demand model with attribute-based screening.
Reminder: there is no closed-form solution for demand, thus we need to integrate not only over the posterior distribution of parameters and the error distribution.
The function outputs a tibble containing id, task, alt, p, attributes, draws from the posterior of demand.
Eerror realisations can be pre-supplied to the epsilon_not. This helps create smooth demand curves or conduct optimization.
vd_dem_vdm_screen(vd, est, epsilon_not = NULL, error_dist = NULL, cores = NULL)vd_dem_vdm_screen(vd, est, epsilon_not = NULL, error_dist = NULL, cores = NULL)
vd |
data |
est |
ec-model draws |
epsilon_not |
(optional) error realizations |
error_dist |
(optional) A string defining the error term distribution (default: 'EV1') |
cores |
(optional) cores |
Draws of expected demand
prep_newprediction() to match vd's factor levels,
ec_gen_err_normal() for pre-generating error realizations and
vd_est_vdm_screen() for estimating the corresponding model
data(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm_screen(R=20, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<20) %>% vd_dem_vdm_screen(icecream_est, cores=2) #column .demdraws contains draws from posterior of predicted demanddata(icecream) #run MCMC sampler (use way more than 20 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<20) %>% vd_est_vdm_screen(R=20, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<20) %>% vd_dem_vdm_screen(icecream_est, cores=2) #column .demdraws contains draws from posterior of predicted demand
Generating demand predictions for volumetric demand model with set-size adjustment.
Reminder: there is no closed-form solution for demand, thus we need to integrate not only over the posterior distribution of parameters and the error distribution.
The function outputs a tibble containing id, task, alt, p, attributes, draws from the posterior of demand.
Eerror realizations can be pre-supplied to the epsilon_not. This helps create smooth demand curves or conduct optimization.
vd_dem_vdm_ss(vd, est, epsilon_not = NULL, cores = NULL)vd_dem_vdm_ss(vd, est, epsilon_not = NULL, cores = NULL)
vd |
data |
est |
ec-model draws |
epsilon_not |
(optional) error realizations |
cores |
(optional) cores |
Draws of expected demand
prep_newprediction() to match vd's factor levels,
ec_gen_err_ev1() for pre-generating error realizations and
vd_est_vdm_ss() for estimating the corresponding model
data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm_ss(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm_ss(icecream_est, cores=2) #column .demdraws contains draws from posterior of predicted demanddata(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<10) %>% vd_est_vdm_ss(R=10, keep=1, cores=2) #Generate demand predictions icecream_predicted_demand= icecream %>% dplyr::filter(id<10) %>% vd_dem_vdm_ss(icecream_est, cores=2) #column .demdraws contains draws from posterior of predicted demand
Estimate volumetric demand model
vd_est_vdm( vd, tidy = TRUE, R = 1e+05, keep = 10, cores = NULL, error_dist = "EV1", control = list(include_data = TRUE) )vd_est_vdm( vd, tidy = TRUE, R = 1e+05, keep = 10, cores = NULL, error_dist = "EV1", control = list(include_data = TRUE) )
vd |
A tibble, containing volumetric demand data (long format) |
tidy |
A logical, whether to apply 'echoice2' tidier function (default: TRUE) |
R |
A numeric, no of draws |
keep |
A numeric, thinning factor |
cores |
An integer, no of CPU cores to use (default: auto-detect) |
error_dist |
A string defining the error term distribution, 'EV1' or 'Normal' |
control |
A list containing additional settings |
An 'echoice2' draw object, in the form of a list
vd_dem_vdm() to generate demand predictions based on this model
vd_est_vdm_screen() to estimate a volumetric demand model with screening
data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=10, cores=2)data(icecream) #run MCMC sampler (use way more than 10 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<50) %>% vd_est_vdm(R=10, cores=2)
Estimate volumetric demand model with attribute-based conjunctive screening
vd_est_vdm_screen( vd, R = 1e+05, keep = 10, cores = NULL, error_dist = "EV1", price_screen = TRUE, control = list(include_data = TRUE) )vd_est_vdm_screen( vd, R = 1e+05, keep = 10, cores = NULL, error_dist = "EV1", price_screen = TRUE, control = list(include_data = TRUE) )
vd |
volumetric demand data (long format) |
R |
draws |
keep |
thinning |
cores |
no of CPU cores to use (default: auto-detect) |
error_dist |
A string defining the error term distribution, 'EV1' or 'Normal' (default: 'EV1') |
price_screen |
A logical, indicating whether price tag screening should be estimated (default: TRUE) |
control |
list containing additional settings |
est ec-draw object (List)
data(icecream) icecream_est <- icecream %>% vd_est_vdm_screen(R=10, cores=2)data(icecream) icecream_est <- icecream %>% vd_est_vdm_screen(R=10, cores=2)
This model REQUIRES variation in choice-set size
vd_est_vdm_ss( vd, order = 1, R = 1e+05, keep = 10, cores = NULL, control = list(include_data = TRUE) )vd_est_vdm_ss( vd, order = 1, R = 1e+05, keep = 10, cores = NULL, control = list(include_data = TRUE) )
vd |
volumetric demand data (long format) with set size variation |
order |
integer, either 1 or 2 (for now), indicating linear or quadratic set-size effect |
R |
draws |
keep |
thinning |
cores |
no of CPU cores to use (default: auto-detect) |
control |
list containing additional settings |
est ec-draw object (List)
data(icecream) #note that for this example dataset, the model is not identified #because the data lacks variation in set size icecream_est <- icecream %>% vd_est_vdm_ss(R=10, cores=2)data(icecream) #note that for this example dataset, the model is not identified #because the data lacks variation in set size icecream_est <- icecream %>% vd_est_vdm_ss(R=10, cores=2)
Log-Likelihood for compensatory volumetric demand model
vd_LL_vdm(draw, vd, fromdraw = 1)vd_LL_vdm(draw, vd, fromdraw = 1)
draw |
A list, 'echoice2' draws object |
vd |
A tibble, tidy choice data (before dummy-coding) |
fromdraw |
An integer, from which draw onwards to compute LL (i.e., excl. burnin) |
N x Draws Matrix of log-Likelihood values
data(icecream) #fit model icecream_est <- icecream %>% vd_est_vdm(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-vd_LL_vdm(icecream_est, icecream, fromdraw = 2) dim(loglls)data(icecream) #fit model icecream_est <- icecream %>% vd_est_vdm(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-vd_LL_vdm(icecream_est, icecream, fromdraw = 2) dim(loglls)
Log-Likelihood for conjunctive-screening volumetric demand model
vd_LL_vdm_screen(draw, vd, fromdraw = 1)vd_LL_vdm_screen(draw, vd, fromdraw = 1)
draw |
A list, 'echoice2' draws object |
vd |
A tibble, tidy choice data (before dummy-coding) |
fromdraw |
An integer, from which draw onwards to compute LL (i.e., excl. burnin) |
N x Draws Matrix of log-Likelihood values
data(icecream) #fit model icecream_est <- icecream %>% filter(id<20) %>% vd_est_vdm_screen(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-vd_LL_vdm_screen(icecream_est, icecream%>% filter(id<20), fromdraw = 2) dim(loglls)data(icecream) #fit model icecream_est <- icecream %>% filter(id<20) %>% vd_est_vdm_screen(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-vd_LL_vdm_screen(icecream_est, icecream%>% filter(id<20), fromdraw = 2) dim(loglls)
Log-Likelihood for volumetric demand model with set-size variation
vd_LL_vdmss(draw, vd, fromdraw = 1)vd_LL_vdmss(draw, vd, fromdraw = 1)
draw |
A list, 'echoice2' draws object |
vd |
A tibble, tidy choice data (before dummy-coding) |
fromdraw |
An integer, from which draw onwards to compute LL (i.e., excl. burnin) |
N x Draws Matrix of log-Likelihood values
data(icecream) #fit model #note: this is just for demo purposes #on this demo dataset, the model is not identified #due to a lack of set size variation icecream_est <- icecream %>% vd_est_vdm_ss(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-vd_LL_vdmss(icecream_est, icecream, fromdraw = 2) #300 respondents, 10 draws dim(loglls)data(icecream) #fit model #note: this is just for demo purposes #on this demo dataset, the model is not identified #due to a lack of set size variation icecream_est <- icecream %>% vd_est_vdm_ss(R=10, keep=1, cores=2) #compute likelihood for each subject in each draw loglls<-vd_LL_vdmss(icecream_est, icecream, fromdraw = 2) #300 respondents, 10 draws dim(loglls)
Generate tidy choice data with dummies from long-format choice data
vd_long_tidy(longdata)vd_long_tidy(longdata)
longdata |
tibble |
tibble
data(icecream) vd_long_tidy(icecream)data(icecream) vd_long_tidy(icecream)
This utility function prepares tidy choice data for fast MCMC samplers.
vd_prepare(dt, Af = NULL)vd_prepare(dt, Af = NULL)
dt |
tidy choice data (columns: id, task, alt, x, p, attributes) |
Af |
(optional) contains a full design matrix (for attribute-based screening), or, more generally, a design matrix used for attribute-based screening |
Note: This function is only exported because it makes it easier to tinker with this package.
This function re-arranges choice data for fast access in highly-optimized MCMC samplers.
It Pre-computes task-wise total expenditures sumpsx and generates indices xfr,xto,lfr,lto for fast data access.
list containing information for estimation functions
#minimal data example dt <- structure(list(id = c(1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L), task = c(1L, 1L, 1L, 2L, 2L, 2L, 1L, 1L, 1L, 2L, 2L, 2L), alt = c(1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L), x = c(1, 0, 2, 1, 0, 1, 2, 3, 1, 1, 0, 1), p = c(0, 1, 1, 1, 2, 0, 2, 2, 1, 2, 1, 1), attr2 = c(1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0), attr1 = c(0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 1, 1)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA,-12L)) #run prep function test <- dt %>% vd_prepare#minimal data example dt <- structure(list(id = c(1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L), task = c(1L, 1L, 1L, 2L, 2L, 2L, 1L, 1L, 1L, 2L, 2L, 2L), alt = c(1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L), x = c(1, 0, 2, 1, 0, 1, 2, 3, 1, 1, 0, 1), p = c(0, 1, 1, 1, 2, 0, 2, 2, 1, 2, 1, 1), attr2 = c(1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0), attr1 = c(0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 1, 1)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA,-12L)) #run prep function test <- dt %>% vd_prepare
This utility function prepares tidy choice data (without x) for fast data access.
vd_prepare_nox(dt, Af = NULL)vd_prepare_nox(dt, Af = NULL)
dt |
tidy choice data (columns: id, task, alt, p, attributes) |
Af |
(optional) contains a full design matrix (for attribute-based screening), or, more generally, a design matrix used for attribute-based screening |
Note: This function is only exported because it makes it easier to tinker with this package. This function re-arranges choice data for fast access, mainly for demand prediction.
list containing information for prediction functions
#Minimal example: #One attribute with 3 levels, 2 subjects, 3 alternatives, 2 tasks dt <- structure(list(id = c(1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L), task = c(1L, 1L, 1L, 2L, 2L, 2L, 1L, 1L, 1L, 2L, 2L, 2L), alt = c(1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L), x = c(1, 0, 2, 1, 0, 1, 2, 3, 1, 1, 0, 1), p = c(0, 1, 1, 1, 2, 0, 2, 2, 1, 2, 1, 1), attr2 = c(1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0), attr1 = c(0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 1, 1)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA,-12L)) test <- dt %>% dplyr::select(-all_of("x")) %>% vd_prepare_nox()#Minimal example: #One attribute with 3 levels, 2 subjects, 3 alternatives, 2 tasks dt <- structure(list(id = c(1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L), task = c(1L, 1L, 1L, 2L, 2L, 2L, 1L, 1L, 1L, 2L, 2L, 2L), alt = c(1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L, 1L, 2L, 3L), x = c(1, 0, 2, 1, 0, 1, 2, 3, 1, 1, 0, 1), p = c(0, 1, 1, 1, 2, 0, 2, 2, 1, 2, 1, 1), attr2 = c(1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0), attr1 = c(0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 1, 1)), class = c("tbl_df", "tbl", "data.frame"), row.names = c(NA,-12L)) test <- dt %>% dplyr::select(-all_of("x")) %>% vd_prepare_nox()
Thin 'echoice2'-vd draw objects
vd_thin_draw(est, burnin_perc = 0.5, total_draws = NULL)vd_thin_draw(est, burnin_perc = 0.5, total_draws = NULL)
est |
is an 'echoice2' draw object (list) |
burnin_perc |
how much burn-in to remove |
total_draws |
how many draws to keep after thinning |
thinned 'echoice2' draw object (list)
data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=10, keep = 1, cores=2) #without thinning, yields R=50 draWs dim(icecream_est$MUDraw) icecream_est_thinned <- vd_thin_draw(icecream_est,.5) #26 draws left after thinning about half dim(icecream_est_thinned$MUDraw)data(icecream) #run MCMC sampler (use way more than 50 draws for actual use) icecream_est <- icecream %>% dplyr::filter(id<100) %>% vd_est_vdm(R=10, keep = 1, cores=2) #without thinning, yields R=50 draWs dim(icecream_est$MUDraw) icecream_est_thinned <- vd_thin_draw(icecream_est,.5) #26 draws left after thinning about half dim(icecream_est_thinned$MUDraw)