915439_en-US

liliconcom

Scripting Help Requested - Making a simulator for Type 1 errors in equivalence testing

Hi All,

 

I'm trying to write a script that will generate a random dataset (n=3 reference and test measurements) and then run a simulation with some number (1000 or whatever).  For each set I'll run an equivalence test and I'd like the output to be a report/table with the frequency of events where equivalence was demonstrated.  The goal of this tool will be to give a probability of type 1 errors given a certain equivalence margin.

I've landed on the following code but this is not working and I'm pretty much stuck.  Any direction/help that the group can provide would be much appreciated.

Thanks!

Chris

// Parameters
numSimulations = 10;
nPerGroup = 3;
refMean = 99.1;
testMean = 98.8;
sd = 0.1;
equivMargin = 0.3;
type1Errors = 0;

// Critical t-value for 90% CI with df = 2n - 2
df = 2 * nPerGroup - 2;
tCrit = StudentTQuantile(0.950, df);

// Create summary table
summaryTable = New Table("Simulation Summary",
    New Column("Simulation", Numeric),
    New Column("CI_Lower", Numeric),
    New Column("CI_Upper", Numeric),
    New Column("Equivalence_Concluded", Character),
    New Column("Mean_Diff", Numeric)
);

// Loop through simulations
For(sim = 1, sim <= numSimulations, sim++,

    // Create new data table
    dt = New Table("Sim" || Char(sim),
        Add Rows(nPerGroup * 2),
        New Column("Group", Character),
        New Column("Value", Numeric)
    );

    // Simulate Reference group
    For(i = 1, i <= nPerGroup, i++,
        dt:Group[i] = "Reference";
        dt:Value[i] = Random Normal(refMean, sd);
    );

    // Simulate Test group
    For(i = nPerGroup + 1, i <= nPerGroup * 2, i++,
        dt:Group[i] = "Test";
        dt:Value[i] = Random Normal(testMean, sd);
    );

    // Extract values
    refVals = dt << Get As Matrix( :Value )[1::nPerGroup];
    testVals = dt << Get As Matrix( :Value )[nPerGroup+1::nPerGroup*2];
    meanDiff = Mean(testVals) - Mean(refVals);

    // Sample variances
    varRef = Variance(refVals);
    varTest = Variance(testVals);

    // Pooled variance
    pooledVar = ((nPerGroup - 1) * varRef + (nPerGroup - 1) * varTest) / df;
    pooledSD = Sqrt(pooledVar);

    // Standard error of difference
    seDiff = pooledSD * Sqrt(2 / nPerGroup);

    // 90% CI
    ciLower = meanDiff - tCrit * seDiff;
    ciUpper = meanDiff + tCrit * seDiff;

    // Check equivalence
    eqResult = "No";
    If(ciLower > -equivMargin & ciUpper < equivMargin,
        type1Errors++;
        eqResult = "Yes";
    );

    // Log results
    summaryTable << Add Rows(1);
    summaryTable:Simulation[sim] = sim;
    summaryTable:CI_Lower[sim] = ciLower;
    summaryTable:CI_Upper[sim] = ciUpper;
    summaryTable:Equivalence_Concluded[sim] = eqResult;
    summaryTable:Mean_Diff[sim] = meanDiff;

    // Close the table to avoid clutter
    dt << Close Window;
);

// Show results
New Window("Type I Error Summary",
    Text Box("Total Simulations: " || Char(numSimulations)),
    Text Box("Type I Errors: " || Char(type1Errors)),
    Text Box("Estimated Type I Error Rate: " || Char(Round(type1Errors / numSimulations * 100, 2)) || "%")
);

Automation and Scripting