# [R] Oddd results when computing confidence intervals

Abraham Mathew mathewanalytics at gmail.com
Fri Dec 2 14:38:43 CET 2016

```I have a vector of values, and have written a function that takes each
value in that vector, generates a normal distribution with that value as
the mean, and then finds the interval at different levels. However, these
intervals don't seem to be right (too narrow).

### CREATE PREDICTION INTERVALS

ensemble_forecast = c(200,600,400,500,200,100,200,600,400,500,200,100)
forecast_for=12

lo_intervals = c()
hi_intervals = c()

create_prediction_intervals <- function(use_forecast = ensemble_forecast,
conf_level = 0.90,
do_jitter = FALSE){

conf.levels1 = paste(round(rep(conf_level, forecast_for/2), 2), "0",
sep="")
conf.levels2 = seq((conf_level+0.02), 0.95, length=forecast_for/2)
all.conf.levels = c(conf.levels1, conf.levels2)
all.conf.levels = as.numeric(as.character(all.conf.levels))
all.conf.levels

# forc_num=1
for(forc_num in 1:length(use_forecast)){
message("Executing forecast number: ", forc_num, " at confidence
level: ", all.conf.levels[forc_num])

value = rnorm(5000, mean=use_forecast[forc_num],
sd=sd(use_forecast))

#t.test(value)\$conf.int
#Rmisc::CI(value, ci=0.99)

low = Rmisc::CI(value, ci=all.conf.levels[forc_num])[]
high = Rmisc::CI(value, ci=all.conf.levels[forc_num])[]

#low = t.test(value, conf.level=all.conf.levels[forc_num])\$conf.int
[]
#high = t.test(value, conf.level=all.conf.levels[forc_num])\$
conf.int[]

lo_intervals.tmp <- c(low)
hi_intervals.tmp <- c(high)

if(do_jitter){
if(length(unique(lo_intervals)) <= 3) lo_intervals.tmp <-
round(jitter(lo_intervals.tmp), 0)
if(length(unique(hi_intervals)) <= 3) hi_intervals.tmp <-
round(jitter(hi_intervals.tmp), 0)
lo_intervals <<- c(lo_intervals, lo_intervals.tmp)
hi_intervals <<- c(hi_intervals, hi_intervals.tmp)
} else {
lo_intervals <<- c(lo_intervals, lo_intervals.tmp)
hi_intervals <<- c(hi_intervals, hi_intervals.tmp)
}
}
}

summary(value)
hist(value)
create_prediction_intervals(ensemble_forecast)

Any ideas on what I'm doing wrong?

--

*Abraham MathewData Ninja and Statistical Modeler*

*Minneapolis, MN720-648-0108 at abmathewksAnalytics_Blog
<https://mathewanalytics.wordpress.com/>*

[[alternative HTML version deleted]]

```