# Pairs Cluster Bootstrapped p-Values For mlogit This software estimates p-values using pairs cluster bootstrapped t-statistics for multinomial logit models (Cameron, Gelbach, and Miller 2008). The data set is repeatedly re-sampled by cluster, a model is estimated, and inference is based on the sampling distribution of the pivotal (t) statistic. ```r cluster.bs.mlogit( mod, dat, cluster, ci.level = 0.95, boot.reps = 1000, cluster.se = TRUE, report = TRUE, prog.bar = TRUE, output.replicates = FALSE, seed = NULL ) ``` ## Arguments - `mod`: A model estimated using `mlogit`. - `dat`: The data set used to estimate `mod`. - `cluster`: A formula of the clustering variable. - `ci.level`: What confidence level should CIs reflect? - `boot.reps`: The number of bootstrap samples to draw. - `cluster.se`: Use clustered standard errors (= TRUE) or ordinary SEs (= FALSE) for bootstrap replicates. - `report`: Should a table of results be printed to the console? - `prog.bar`: Show a progress bar of the bootstrap (= TRUE) or not (= FALSE). - `output.replicates`: Should the cluster bootstrap coefficient replicates be output (= TRUE) or not (= FALSE)? - `seed`: Random number seed for replicability (default is NULL). ## Returns A list with the elements - **p.values**: A matrix of the estimated p-values. - **ci**: A matrix of confidence intervals. ## Note Code to estimate GLM clustered standard errors by Mahmood Arai: http://thetarzan.wordpress.com/2011/06/11/clustered-standard-errors-in-r/, although modified slightly to work for `mlogit` models. Cluster SE degrees of freedom correction = (M/(M-1)) with M = the number of clusters. ## Examples ```r ## Not run: ####################################### # example one: train ticket selection ####################################### require(mlogit) data("Train", package="mlogit") Train$choiceid <- 1:nrow(Train) Tr <- dfidx(Train, shape = "wide", varying = 4:11, sep = "_", choice = "choice", idx = list(c("choiceid", "id")), idnames = c(NA, "alt")) Tr$price <- Tr$price/100 * 2.20371 Tr$time <- Tr$time/60 ml.Train <- mlogit(choice ~ price + time + change + comfort | -1, Tr) # compute pairs cluster bootstrapped p-values # note: few reps to speed up example cluster.bs.tr <- cluster.bs.mlogit(ml.Train, Tr, ~ id, boot.reps=100) ################################################################## # example two: predict type of heating system installed in house ################################################################## require(mlogit) data("Heating", package = "mlogit") H <- Heating H$region <- as.numeric(H$region) H.ml <- dfidx(H, shape="wide", choice="depvar", varying=c(3:12), idx = list(c("idcase", "region"))) m <- mlogit(depvar~ic+oc, H.ml) # compute pairs cluster bootstrapped p-values cluster.bs.h <- cluster.bs.mlogit(m, H.ml, ~ region, boot.reps=1000) ## End(Not run) ``` ## References Esarey, Justin, and Andrew Menger. 2017. "Practical and Effective Approaches to Dealing with Clustered Data." **Political Science Research and Methods** forthcoming: 1-35. . Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. 2008. "Bootstrap-Based Improvements for Inference with Clustered Errors." **The Review of Economics and Statistics** 90(3): 414-427. . ## Author(s) Justin Esarey

cluster.bs.mlogit function