Exploration and description

A missing value represents a value that could not be recorded. For several of the variables, such as the mineral determinations, there was no plant material available for analysis. These observations are clearly missing. However, when a variable reflects a zero response (e.g. dry corm weight or number of corms) such data, on reflection, would have been more meaningfully recorded as zero.

Growth was also poor for the 3rd planting date and so one might decide just to analyse the data for the first two plantings. However, for the purposes of this case study, we shall just omit the 4th planting date with its many zeros. This can be achieved by Spread → Restrict/Filter → To groups (factor levels) ... and selecting levels 1-3 for Planting_date.

Before proceeding with the statistical analysis of the data we should just check for any extreme data values. By carrying out a preliminary analysis of variance (Stats → Analysis of Variance ...) and selecting 'Split-plot Design' (see alongside) we find the message below contained within the output. These residual values are over four times their standard errors and hence outliers.

The entries on the recording sheets matched the values entered into the spreadsheet and no explanation could be given for these unusual values. They were therefore replaced by the * missing value code and a copy of this sheet saved in a separate spreadsheet 'Edited data' in CS16Data1a.