Comparing two designs (or anything else!) using paired sample T-tests - Improving the User Experience through Practical Data Analytics - page 88

FIGURE 3.9
Output when data incorrectly treated as independent samples; Excel.
The bottom line is that the variability in weights of the individuals before the
intervention should not matter if it is paired data, while it should (and indeed, does—
in a big way!) when having independent samples.
Fortunately, we don't have to really worry about this detail. All we need to do
is to correctly tell Excel or SPSS whether we have independent samples or paired
samples—pretty nifty, right 2 ?
3.7 SUMMARY
In this chapter we have presented the t-test for testing equality of means using paired
data. We described what paired data indicates, how to instruct Excel and SPSS that paired
data are being analyzed, illustrated the paired-data t-test in both Excel and SPSS, and
2 By the way, we have not mentioned inding a conidence interval in the context of having paired data.
However, inding a meaningful conidence interval and error bar for paired data can be done. It would
require irst subtracting one column from another, row by row, and inding a single conidence interval
and error bar on the resulting difference column.