The {logitr} package requires that data be structured in a
data.frame and arranged in a “long” format [@Wickham2014] where each row contains data on a
single alternative from a choice observation. The choice observations do
not have to be symmetric, meaning they can have a “ragged” structure
where different choice observations have different numbers of
alternatives. The data must also include variables for each of the
following:
1 is chosen, 0
is not chosen). Only one alternative should have a 1 per
choice observation.obsID variable would be
1, 1, 2, 2, 3, 3.The “Data Formatting and Encoding” vignette has more details about the required data format.
Models are specified and estimated using the logitr()
function. The data argument should be set to the data frame
containing the data, and the outcome and obsID
arguments should be set to the column names in the data frame that
correspond to the dummy-coded outcome (choice) variable and the
observation ID variable, respectively. All variables to be used as model
covariates should be provided as a vector of column names to the
pars argument. Each variable in the vector is additively
included as a covariate in the utility model, with the interpretation
that they represent utilities in preference space models and WTPs in a
WTP space model.
For example, consider a preference space model where the utility for yogurt is given by the following utility model:
\[\begin{equation} u_{j} = \alpha p_{j} + \beta_1 x_{j1} + \beta_2 x_{j2} + \beta_3 x_{j3} + \beta_4 x_{j4} + \varepsilon_{j}, \label{eq:yogurtUtilityPref} \end{equation}\]
where \(p_{j}\) is
price, \(x_{j1}\) is
feat, and \(x_{j2-4}\) are
dummy-coded variables for each brand (with the fourth brand
representing the reference level). This model can be estimated using the
logitr() function as follows:
library("logitr")
mnl_pref <- logitr(
data = yogurt,
outcome = "choice",
obsID = "obsID",
pars = c("price", "feat", "brand")
)The equivalent model in the WTP space is given by the following utility model:
\[\begin{equation} u_{j} = \lambda \left( \omega_1 x_{j1} + \omega_1 x_{j2} + \omega_1 x_{j3} + \omega_2 x_{j4} - p_{j} \right) + \varepsilon_{j}, \label{eq:yogurtUtilityWtp} \end{equation}\]
To specify this model, simply move "price" from the
pars argument to the scalePar argument:
mnl_wtp <- logitr(
data = yogurt,
outcome = "choice",
obsID = "obsID",
pars = c("feat", "brand"),
scalePar = "price"
) In the above model, the variables in pars are marginal
WTPs, whereas in the preference space model they are marginal utilities.
Price is separately specified with the scalePar argument
because it acts as a scaling term in WTP space models. While price is
the most typical scaling variable, other continuous variables can also
be used, such as time.
Interactions between covariates can be entered in the
pars vector separated by the * symbol. For
example, an interaction between price with
feat in a preference space model could be included by
specifying
pars = c("price", "feat", "brand", "price*feat"), or even
more concisely just pars = c("price*feat", "brand") as the
interaction between price and feat will
produce individual parameters for price and
feat in addition to the interaction parameter.
Both of these examples are multinomial logit models with fixed parameters. See the “Estimating Multinomial Logit Models” vignette for more details.
Since WTP space models are non-linear and have non-convex
log-likelihood functions, it is recommended to use a multi-start search
to run the optimization loop multiple times to search for different
local minima. This is implemented using the numMultiStarts
argument, e.g.:
mnl_wtp <- logitr(
data = yogurt,
outcome = "choice",
obsID = "obsID",
pars = c("feat", "brand"),
scalePar = "price",
numMultiStarts = 10
)The multi-start is parallelized by default for faster estimation, and
the number of cores to use can be manually set using the
numCores argument. If numCores is not provide,
then the number of cores is set to
parallel::detectCores() - 1. For models with larger data
sets, you may need to set numCores = 1 to avoid memory
overflow issues.
To estimate a mixed logit model, use the randPars
argument in the logitr() function to denote which
parameters will be modeled with a distribution. The current package
version supports normal ("n"), log-normal
("ln"), and zero-censored normal ("cn")
distributions.
For example, assume the observed utility for yogurts was \(v_{j} = \alpha p_{j} + \beta_1 x_{j1} + \beta_2
x_{j2} + \beta_3 x_{j3} + \beta_4 x_{j4}\), where \(p_{j}\) is price, \(x_{j1}\) is feat, and \(x_{j2-4}\) are dummy-coded variables for
brand. To model feat as well as each of the
brands as normally-distributed, set
randPars = c(feat = "n", brand = "n"):
mxl_pref <- logitr(
data = yogurt,
outcome = 'choice',
obsID = 'obsID',
pars = c('price', 'feat', 'brand'),
randPars = c(feat = 'n', brand = 'n'),
numMultiStarts = 10
)Since mixed logit models also have non-convex log-likelihood functions, it is recommended to use a multi-start search to run the optimization loop multiple times to search for different local minima.
See the “Estimating Mixed Logit Models” vignette for more details.
Use the summary() function to print a summary of the
results from an estimated model, e.g.
summary(mnl_pref)
#> =================================================
#>
#> Model estimated on: Thu Jun 16 17:36:36 2022
#>
#> Using logitr version: 0.7.0
#>
#> Call:
#> logitr(data = yogurt, outcome = "choice", obsID = "obsID", pars = c("price",
#> "feat", "brand"))
#>
#> Frequencies of alternatives:
#> 1 2 3 4
#> 0.402156 0.029436 0.229270 0.339138
#>
#> Exit Status: 3, Optimization stopped because ftol_rel or ftol_abs was reached.
#>
#> Model Type: Multinomial Logit
#> Model Space: Preference
#> Model Run: 1 of 1
#> Iterations: 21
#> Elapsed Time: 0h:0m:0.04s
#> Algorithm: NLOPT_LD_LBFGS
#> Weights Used?: FALSE
#> Robust? FALSE
#>
#> Model Coefficients:
#> Estimate Std. Error z-value Pr(>|z|)
#> price -0.366555 0.024365 -15.0441 < 2.2e-16 ***
#> feat 0.491439 0.120062 4.0932 4.254e-05 ***
#> brandhiland -3.715477 0.145417 -25.5506 < 2.2e-16 ***
#> brandweight -0.641138 0.054498 -11.7645 < 2.2e-16 ***
#> brandyoplait 0.734519 0.080642 9.1084 < 2.2e-16 ***
#> ---
#> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#>
#> Log-Likelihood: -2656.8878790
#> Null Log-Likelihood: -3343.7419990
#> AIC: 5323.7757580
#> BIC: 5352.7168000
#> McFadden R2: 0.2054148
#> Adj McFadden R2: 0.2039195
#> Number of Observations: 2412.0000000Use statusCodes() to print a description of each status
code from the nloptr optimization routine.
You can also extract other values of interest at the solution, such as:
The estimated coefficients
coef(mnl_pref)
#> price feat brandhiland brandweight brandyoplait
#> -0.3665546 0.4914392 -3.7154773 -0.6411384 0.7345195The coefficient standard errors
se(mnl_pref)
#> price feat brandhiland brandweight brandyoplait
#> 0.02436526 0.12006175 0.14541671 0.05449794 0.08064229The log-likelihood
logLik(mnl_pref)
#> 'log Lik.' -2656.888 (df=5)The variance-covariance matrix
vcov(mnl_pref)
#> price feat brandhiland brandweight
#> price 0.0005936657 5.729619e-04 0.001851795 1.249988e-04
#> feat 0.0005729619 1.441482e-02 0.000855011 5.092011e-06
#> brandhiland 0.0018517954 8.550110e-04 0.021146019 1.490080e-03
#> brandweight 0.0001249988 5.092011e-06 0.001490080 2.970026e-03
#> brandyoplait -0.0015377721 -1.821331e-03 -0.003681036 7.779428e-04
#> brandyoplait
#> price -0.0015377721
#> feat -0.0018213311
#> brandhiland -0.0036810363
#> brandweight 0.0007779427
#> brandyoplait 0.0065031782For models in the preference space, a summary table of the computed
WTP based on the estimated preference space parameters can be obtained
with the wtp() function. For example, the computed WTP from
the previously estimated fixed parameter model can be obtained with the
following command:
wtp(mnl_pref, scalePar = "price")
#> Estimate Std. Error z-value Pr(>|z|)
#> scalePar 0.366555 0.024467 14.9817 < 2.2e-16 ***
#> feat 1.340699 0.358947 3.7351 0.0001877 ***
#> brandhiland -10.136219 0.585208 -17.3207 < 2.2e-16 ***
#> brandweight -1.749094 0.181953 -9.6129 < 2.2e-16 ***
#> brandyoplait 2.003848 0.143351 13.9786 < 2.2e-16 ***
#> ---
#> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1The wtp() function divides the non-price parameters by
the negative of the scalePar parameter (usually “price”).
Standard errors are estimated using the Krinsky and Robb parametric
bootstrapping method [@Krinsky1986].
Similarly, the wtpCompare() function can be used to compare
the WTP from a WTP space model with that computed from an equivalent
preference space model:
wtpCompare(mnl_pref, mnl_wtp, scalePar = "price")
#> pref wtp difference
#> scalePar 0.3665546 0.3665832 0.00002867
#> feat 1.3406987 1.3405926 -0.00010605
#> brandhiland -10.1362190 -10.1357635 0.00045548
#> brandweight -1.7490940 -1.7490826 0.00001133
#> brandyoplait 2.0038476 2.0038208 -0.00002686
#> logLik -2656.8878790 -2656.8878779 0.00000106Estimated models can be used to predict probabilities and outcomes
for a set (or multiple sets) of alternatives based on an estimated
model. As an example, consider predicting probabilities for two of the
choice observations from the yogurt dataset:
data <- subset(
yogurt, obsID %in% c(42, 13),
select = c('obsID', 'alt', 'choice', 'price', 'feat', 'brand')
)
data
#> obsID alt choice price feat brand
#> 49 13 1 0 8.1 0 dannon
#> 50 13 2 0 5.0 0 hiland
#> 51 13 3 1 8.6 0 weight
#> 52 13 4 0 10.8 0 yoplait
#> 165 42 1 1 6.3 0 dannon
#> 166 42 2 0 6.1 1 hiland
#> 167 42 3 0 7.9 0 weight
#> 168 42 4 0 11.5 0 yoplaitIn the example below, the probabilities for these two sets of
alternatives are computed using the fixed parameter
mnl_pref model using the predict() method:
probs <- predict(
mnl_pref,
newdata = data,
obsID = "obsID",
ci = 0.95
)
probs
#> obsID predicted_prob predicted_prob_lower predicted_prob_upper
#> 49 13 0.43685145 0.41656557 0.45783129
#> 50 13 0.03312986 0.02610886 0.04178415
#> 51 13 0.19155548 0.17628955 0.20749984
#> 52 13 0.33846321 0.32068460 0.35868268
#> 165 42 0.60764778 0.57673139 0.63854665
#> 166 42 0.02602007 0.01836277 0.03686672
#> 167 42 0.17803313 0.16198270 0.19363903
#> 168 42 0.18829902 0.16913650 0.20886676The resulting probs data frame contains the expected
probabilities for each alternative. The lower and upper predictions
reflect a 95% confidence interval (controlled by the ci
argument), which are estimated using the Krinsky and Robb parametric
bootstrapping method [@Krinsky1986]. The
default is ci = NULL, in which case no CI predictions are
made.
WTP space models can also be used to predict probabilities:
probs <- predict(
mnl_wtp,
newdata = data,
obsID = "obsID",
ci = 0.95
)
probs
#> obsID predicted_prob predicted_prob_lower predicted_prob_upper
#> 49 13 0.43686141 0.41521460 0.45738047
#> 50 13 0.03312947 0.02660889 0.04215556
#> 51 13 0.19154829 0.17619478 0.20745819
#> 52 13 0.33846083 0.31836548 0.35837208
#> 165 42 0.60767120 0.57302344 0.63676998
#> 166 42 0.02601800 0.01880420 0.03584319
#> 167 42 0.17802363 0.16202434 0.19488103
#> 168 42 0.18828717 0.16920672 0.20904598You can also use the predict() method to predict
outcomes by setting type = "outcome" (the default value is
"prob" for predicting probabilities). If no new data are
provided for newdata, then outcomes will be predicted for
every alternative in the original data used to estimate the model. In
the example below the returnData argument is also set to
TRUE so that the predicted outcomes can be compared to the
actual ones.
outcomes <- predict(
mnl_pref,
type = "outcome",
returnData = TRUE
)
head(outcomes[c('obsID', 'choice', 'predicted_outcome')])
#> obsID choice predicted_outcome
#> 1 1 0 1
#> 2 1 0 0
#> 3 1 1 0
#> 4 1 0 0
#> 5 2 1 1
#> 6 2 0 0You can quickly compute the accuracy by dividing the number of correctly predicted choices by the total number of choices:
chosen <- subset(outcomes, choice == 1)
chosen$correct <- chosen$choice == chosen$predicted_outcome
sum(chosen$correct) / nrow(chosen)
#> [1] 0.3677446See the “Predicting Probabilities and Choices from Estimated Models” vignette for more details about making predictions.