Influence

Influence Definition

Influence of a variate yu on the population attribute when a variate yu is removed from the population is measured by the following expression:

Δ(α,u) = α(y1, …, yu − 1, yu, yu + 1, …, yN) − α(y1, …, yu − 1, yu + 1, …, yN) for each unit u in the population, and α is an arbitrary population attribute. Note the first part of the equation contains the unit u, and the unit u is removed in the second part of the equation.

Influence Interpretation

Ideally, each unit should have the same influence. However this is often not the case. If a unit has different unit than the rest of the population, then this unit could be an error, or, this unit can be an interesting unit to look into to determine why this unit has larger/smaller impact than the rest of the units when removed from the population.

Influence Examples

We can find the influence with calculations by hand or use R. Below, an example of finding influence on the attribute population mean is showed.

Hand Calculation

For the population mean attribute, the population mean without unit u can be written as \(\\begin{aligned}\\alpha(y\_1,\\ldots,y\_{u-1},y\_{u+1},\\ldots,y\_N) = \\frac{1}{N-1} \\sum\_{k\\in\\mathcal{P},k \\neq u}y\_k = \\frac{\\sum\_{k\\in\\mathcal{P}}{y\_k} - y\_u}{N -1}= \\frac{N \\bar{y}-y\_u}{N -1}\\end{aligned}\) So, the influence Δ(α, u) for the population mean is $\begin{aligned} \Delta(\alpha,u) &= \alpha(y_1,\ldots,y_{u-1},y_u,y_{u+1},\ldots,y_N) - \alpha(y_1,\ldots,y_{u-1},y_{u+1},\ldots,y_N) \
&= \bar{y} - \frac{N \bar{y}-y_u}{N -1} = \frac{N \bar{y} - \bar{y}- N \bar{y}+y_u}{N -1} =\frac {y_u- \bar{y}}{N -1} \end{aligned}$ When we encounter a word problem, we just need to plug in the values to $\Delta(\alpha,u)=\frac {y_u- \bar{y}}{N -1}$ to find the influence for the given u for population mean.

Calculation and Plotting using R

There are three main ways to approach this problem using R.

  1. Looping through each value to calculate Δ.
  2. Create a matrix and use the apply function.
  3. Summing numeric vectors. We will use Approach 3 for this example. Refer to the course notes on 2.2.3 to see all the approaches.
# Setup
directory <-"/Users/chang/OneDrive - University of Waterloo/3A/STAT 341/a2"
dirsep <-"/"
filename <-paste(directory,"agpop_data.csv",sep=dirsep)
agpop <-read.csv(filename,header=TRUE)

After setting up, the following is approach 3 for calculating inference.

y = agpop$farms87
ybar = mean(y)

# Note that y is a vector and we calculated inference of every value here
delta = (y-ybar)/(length(y)-1)

Normally, we would only plug in the value for one yu. The above R calculation finds inference for all yu for plotting.

Now, let’s plot the influence graph for each unit u with the following R code,

par(mfrow =c(1,2))

plot(delta,main ="Influence for Average",pch =19,
     col =adjustcolor("black",alpha =0.2),
     xlab ="Index",ylab =bquote(Delta))

plot(y, delta,main ="Influence for Average",pch =19,
     col =adjustcolor("black",alpha =0.2),
     xlab ="Farm Size (y)",ylab =bquote(Delta))

Result Interpretations

In this example, we could see that there are a couple of large farms that have much more influence than other units to population mean.

# Gives unit number for the largest values
which(delta > 1.5)

## [1] 172 199 216

So, unit 172 (Fresno), 199(San Diego), and 216 (Tulare) have the largest influences. These three farms are also the largest in size.

The three largest farms have the most influence (i.e. the three most extreme values on the right plot). It makes sense for the three largest farms to have the biggest influence in this case because large values would impact population mean more.

In general, while interpreting the results for influence, it’s useful to plot graphs for each unit u using R. Then, we should determine the units with the largest influence on the specific population attribute and find out why. The influence plot often provides useful insights on the population attribute.