NaN's appearing in sdsm under "model = chi2" with certain dataset

[domagal9]

Using the gawc11 dataset, NaN's appearing in sdsm$positive and sdsm$negative when "model = chi2" is used. gawc11.zip

Issue replicated on mac and windows Not necessarily when there are exact 0's & 1's cause can occur under other models (e.g. logit, polytope). Does not occur in other datasets (e.g. senate)

Rep code: data <- read.csv("gawc11.csv", row.names = 1) data <- as.matrix(data) sdsmchi2 <- sdsm(data, model="chi2") sdsmchi2$positive sdsmchi2$negative

Error is specifically occurring in the rna function under the command pp = rep(pp, wts) The error is "rep(pp, wts) : invalid 'times' argument". This command does not have an error with other datasets.

[domagal9]

Reason: This problem is occurring within the RNA function The part that is causing the error is the line: ind = gammak/(6*sigmak^3) This is causing an error because it is dividing by zero. sigmak is zero because each value of pp is zero

Why is each value of pp zero?

• pp is a vector
• pp = prob.mat[i,] * prob.mat[j,]
• When either of these rows (e.g. prob.mat[i,]) contains all zeros, pp becomes a vector of zeros
• Why are these zero? The chi2 test is defining them as zero
• They are occurring when a row or column has a zero marginal
• Thus the chi2 value rowmarg*colmarg / totalsum = 0 for that entry

This is good! This means it is directly because of the dataset, not an error. We could remedy this by (1) Adding a statement that all zero bipartite data shouldn’t be used or (2) Adding a super small amount to the chi2 values with the line: probs <- probs+0.000000000000001