Auto-regressive app
ar_app()This shiny app allows the user to explore the effect of autocorrelation on regression models if the autocorrelation is ignored. The app simulates and plots a first order autoregressive process, a time series where the next value depends only the previous value. The strength of the autoregressive coefficient can be changed with a slider to values strictly between -1 and 1. Positive values generate positive autocorrelation (neighbouring values are more similar than expected); negative values generate negative autocorrelation (neighbouring values are less similar than expected).
A linear regression model can be added to the plot. The model assumes that there is no autocorrelation. If the autocorrelation is high, the slope is often quite different from zero.
In the second tab, the results of 100 linear models fitted to autocorrelated time series and 100 models fitted to time series with independent observations are shown. A histogram shows the distribution of p-values for the models fitted to autocorrelated time series. If there was not autocorrelation, the histogram is expected to be flat.
if (FALSE) { # \dontrun{
ar_app()
} # }