Jacknife analysis

[#1]

Hi,

I am interested to do Jacknife analysis (Leave - One - Out) to get an an estimate of error in the mean of a sample distribution. I saw Bootstrap in JMP Pro and appreciate if anyone has done Jacknife analysis in JMP.  Thanks.

[#2]

What platform are you working in?

There are some platforms that have Leave-one-out as an option for cross-validation.  However, if you don't see it as an option you can select K-fold cross-validation and set the K value to the number of rows you have in your data table.

[#3]

Thanks for your suggestions. I have JMP PRO 14.

[#4]

Jackknife is a bit different from bootstrap, as I'm sure you know, but there is some jackknife distance stuff for outlier analysis in the multivariate platform.  That's the only place I know of with any built-in jackknife analysis.

https://www.jmp.com/support/help/14/distance-measures.shtml#240680

For your use case, you would probably need to script it up by hand.  It wouldn't be hard, although, why not bootstrap if you have access to JMP Pro.  I thought of a way to vectorize the operation for a standard arithmetic mean.  Here's an example.

//Create Example Table
n = 100; //number of observations
dt = New Table("Jackknife Example",<< New Column("X",Formula(Random Normal(100,10))));
dt << Add Rows(n);

//Jackknife Calculations
b = Identity(n)*-1 + 1; //matrix of 1s with 0s on the diagonal
Xvals = dt:X << Get Values; //get column X as vector
repX = shape(Xvals, n, n); //repeat vector X n times in new matrix (repeated as row vectors)

means_jack = repX:*b*j(n,1,1)/(n-1); //compute jackknife means

//Results
xbar = mean(means_jack);
var_jack = (n-1)^2/n*stddev(means_jack)^2;
se_jack = sqrt(var_jack);

[#5]

Thanks Cameron for your help and script. I do have JMP Pro 14 and certainly can use Bootstrap as well.