The function performNB calculates the posterior probabilities of a dichotomous class
variable given a set of covariates using Bayes rule and either a univariate (default, orType = "univariate"
odds ratio or a bootstrapped adjusted odds ratio via logistic regression.
performNB(
training,
prediction,
obsIDVar,
goldStdVar,
covariates,
l = 1,
orType = "univariate",
nBS = 100,
pSampled = 1
)The training dataset name.
The prediction dataset name.
The variable name (in quotes) of the observation ID variable.
The variable name (in quotes) of the outcome in the training dataset
(needs to be a logical variable with value TRUE for observations with
the outcome of interest.)
A character vector containing the covariate variable names. All covariates need to be categorical factor variables.
Laplace smoothing parameter that is added to each cell (a value of 0 indicates no smoothing).
Takes value "univariate" or "adjusted". "univariate" produces contingency table
odds ratios and "adjusted" produces adjusted odds ratios from a bootstrapped multivariable logistic regression.
Number of bootstrap samples to run in each cross-validation fold/iteration (default is 100). Only
relevant when orType = "adjusted".
Proportion of unlinked cases to include in bootstrap sample (default is 1, i.e.a true
bootstrap). Only relevant when orType = "adjusted".
List containing two dataframes:
probabilities - a dataframe combining training and prediction
with predictied probabilities for the prediction dataframe. Column names:
<obsIDVar> - the observation ID with the name specified
p - the probability that <goldStdVar> = TRUE for observations in the
prediction dataset.
estimates - a dataframe with the effect estimates derived from the training dataset.
Column names:
level - the covariate name and level
est - the log odds ratio for this covariate and level
se - the standard error of the log odds ratio
The main purpose of this function is to be used by nbProbabilities to
estimate the relative transmission probability between individuals in an infectious
disease outbreak. However, it can be used more generally to estimate the probability
of any dichotomous outcome given a set of categorical covariates and adjusted odds ratios of such
dichotomous outcome.
This function also generates odds ratios describing the associations between covariates in the training data
and outcome defined in the gold standard variable (goldStdVar) argument. Unadjusted odds ratios are the default.
These odds ratios are produced using contingency table methods. Adjusted odds ratios are calculated via bootstrapped
logistic regression to produce non-parametric standard errors. The bootstrap is controlled by parameters nBS,
the number of bootstrap samples to run, and pSampled, the proportion of unlinked cases to include in the bootstrap
sample. pSampled is recommended only for large datasets in which it is computationally unfeasible to run a full
bootstrap. Sensitivity analyses should be run to determine an adequate value for pSampled.
The function needs a training dataset with the outcome variable (goldStdVar)
which is TRUE for those who have the value of interest and FALSE
for those who do not. The probability of having the outcome
(<goldStdVar> = TRUE) is predicted in the prediction dataset.
## Use iris dataset and predict if a flower is of the specices "virginica".
data(iris)
irisNew <- iris
## Creating an id variable
irisNew$id <- seq(1:nrow(irisNew))
## Creating logical variable indicating if the flower is of the species virginica
irisNew$spVirginica <- irisNew$Species == "virginica"
## Creating categorical/factor versions of the covariates
irisNew$Sepal.Length.Cat <- factor(cut(irisNew$Sepal.Length, c(0, 5, 6, 7, Inf)),
labels = c("<=5.0", "5.1-6.0", "6.1-7.0", "7.1+"))
irisNew$Sepal.Width.Cat <- factor(cut(irisNew$Sepal.Width, c(0, 2.5, 3, 3.5, Inf)),
labels = c("<=2.5", "2.6-3.0", "3.1-3.5", "3.6+"))
irisNew$Petal.Length.Cat <- factor(cut(irisNew$Petal.Length, c(0, 2, 4, 6, Inf)),
labels = c("<=2.0", "2.1-4.0", "4.1-6.0", "6.0+"))
irisNew$Petal.Width.Cat <- factor(cut(irisNew$Petal.Width, c(0, 1, 2, Inf)),
labels = c("<=1.0", "1.1-2.0", "2.1+"))
## Using NB to predict if the species is virginica
## (training and predicting on same dataset)
pred <- performNB(irisNew, irisNew, obsIDVar = "id",
goldStdVar = "spVirginica",
covariates = c("Sepal.Length.Cat", "Sepal.Width.Cat",
"Petal.Length.Cat", "Petal.Width.Cat"), l = 1)
irisResults <- merge(irisNew, pred$probabilities, by = "id")
tapply(irisResults$p, irisResults$Species, summary)
#> $setosa
#> Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
#> 0.00003 0.00006 0.00010 0.00010 0.00012 0.00018 50
#>
#> $versicolor
#> Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
#> 0.00018 0.03570 0.43028 0.43323 0.76622 0.85702 50
#>
#> $virginica
#> Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
#> 0.1180 0.8570 0.9920 0.8474 0.9956 0.9999 50
#>