Yates analysis is a method of analyzing data from full and partial factorial design experiments. It requires two levels for each factor in the experiment – these are commonly referred to as ‘high’ and ‘low’ levels, and are represented by + and – signs respectively.
Yates order
‘Yates order’ refers to a specific arrangement of experiment data needed for Yates analysis. To get your data into Yates order, experiments should be arranged so that, with k number of factors, each factor in its own column, 2(k-1) minus signs in a column (low levels) should be followed by the same number of plus signs (high levels), as a pattern down the height of the column. It sounds complicated, but let’s look at a simple example with 3 factors, already in Yates order:
| Trial | Factor 1 | Factor 2 | Factor 3 | Data |
|---|---|---|---|---|
| 1 | – | – | – | |
| 2 | + | – | – | |
| 3 | – | + | – | |
| 4 | + | + | – | |
| 5 | – | – | + | |
| 6 | + | – | + | |
| 7 | – | + | + | |
| 8 | + | + | + |
For the first column in the table above, k = 1. According to Yates order, there should be a pattern of 2(1-1) minus signs followed by 2(1-1) plus signs in this column – or 1 minus sign and 1 plus sign. A quick look at the table will tell you this is the case. For the second factor in the table, Yates order dictates a pattern of 2(2-1) minus signs followed by 2(2-1) plus signs.
Yates analysis output
Most Yates analysis is performed using analysis software. The output consists of:
- Factor identifier. The notation varies depending on the software, but it generally looks like the notation we talk about in the Partial/Fractional Factorial Design topic. Sometimes you’ll see numbers used rather than letters.
- Ranked factor list. This uses least squares to determine the most significant factors – ie, the ones that had most effect on the results. For each factor,
- t-value for each factor.
- residual standard deviation for that factor alone.
- cumulative residual standard deviation for factors up to and including that factor.
