Split plot design - Oats

Introduction

The first example involves the analysis of a split plot design originally presented by Yates (1935). The experiment was conducted to assess the effects on yield of three oat varieties (Golden Rain, Marvellous and Victory) with four levels of nitrogen application (0, 0.2, 0.4 and 0.6 cwt/acre). The field layout consisted of six blocks (labelled I, II, III, IV, V and VI) with three whole-plots per block, each split into four sub-plots. The three varieties were randomly allocated to the three whole-plots while the four levels of nitrogen application were randomly assigned to the four sub-plots within each whole-plot. The data is
                  nitrogen
 block  variety   0.0cwt  0.2cwt  0.4cwt  0.6cwt
           GR      111     130     157     174
   I       M       117     114     161     141
           V       105     140     118     156

           GR       61      91      97     100
   II      M        70     108     126     149
           V        96     124     121     144

           GR       68      64     112      86
   III     M        60     102      89      96
           V        89     129     132     124

           GR       74      89      81     122
   IV      M        64     103     132     133
           V        70      89     104     117

           GR       62      90     100     116
   V       M        80      82      94     126
           V        63      70     109      99

           GR       53      74     118     113
   VI      M        89      82      86     104
           V        97      99     119     121
A standard analysis of these data recognises the two basic elements inherent in the experiment. These two aspects are firstly the stratification of the experiment units, that is the blocks, whole-plots and sub-plots, and secondly, the treatment structure that is superimposed on the experimental material. The latter is of prime interest, in the presence of stratification. Thus the aim of the analysis is to examine the importance of the treatment effects while accounting for the stratification and restricted randomisation of the treatments to the experimental units. The ASReml input file is presented below.
 split plot example
  blocks   6     # Coded 1...6 in first data field of oats.asd
  nitrogen !A 4  # Coded alphabetically
  subplots *     # Coded 1...4
  variety  !A 3  # Coded alphabetically
  wplots   *     # Coded 1...3
  yield
 oats.asd !SKIP 2

 yield ~ mu variety nitrogen variety.nitrogen !r blocks blocks.wplots
 predict nitrogen   # Print table of predicted nitrogen means
 predict variety
 predict variety nitrogen !SED
The data fields were blocks, wplots, subplots, variety, nitrogen and yield. The first five variables are factors that describe the stratification or experiment design and treatments. The standard split plot analysis is achieved by fitting the model terms blocks and blocks.wplots as random effects. The blocks.wplots.subplots term is not listed in the model because this interaction corresponds to the experimental units and is automatically included as the residual term. The fixed effects include the main effects of both variety and nitrogen and their interaction. The tables of predicted means and associated standard errors of differences ( SEDs) have been requested. These are reported in the .pvs file. Abbreviated output is shown below.
           - - - Results from analysis of yield - - -

           Approximate stratum variance decomposition
  Stratum     Degrees-Freedom   Variance      Component Coefficients
  blocks                 5.00    3175.06        12.0     4.0     1.0
  blocks.wplots         10.00    601.331         0.0     4.0     1.0
  Residual Variance     45.00    177.083         0.0     0.0     1.0

  Source                Model  terms     Gamma     Component    Comp/SE   % C
  blocks                    6      6   1.21116       214.477       1.27   0 P
  blocks.wplots            18     18  0.598937       106.062       1.56   0 P
  Variance                 72     60   1.00000       177.083       4.74   0 P

                                    Wald F statistics
      Source of Variation           NumDF     DenDF    F-inc             Prob
    7 mu                                1       5.0   245.14            <.001
    4 variety                           2      10.0     1.49            0.272
    2 nitrogen                          3      45.0    37.69            <.001
    8 variety.nitrogen                  6      45.0     0.30            0.932
For simple variance component models such as the above, the default parameterisation for the variance component parameters is as the ratio to the residual variance. Thus ASReml prints the variance component ratio and variance component for each term in the random model in the columns labelled Gamma and Component respectively. The analysis of variance ( ANOVA) is printed below this summary. The usual decomposition has three strata, with treatment effects separating into different strata as a consequence of the balanced design and the allocation of variety to whole-plots. In this balanced case, it is straightforward to derive the ANOVA estimates of the stratum variances from the REML estimates of the variance components. That is
blocks = 12 S2b + 4S2w + S2e = 3175.1
blocks.wplots = 4S2w + S2e = 601.3
residual =S2e = 177.1

The default output for testing fixed effects used by ASReml is a table of so-called incremental F-statistics. The statistics are simply the appropriate Wald test statistics divided by the number of estimable effects for that term. In this example there are four terms included in the summary. The overall mean (denoted by mu) is of no interest for these data. The tests are sequential, that is the effect of each term is assessed by the change in sums of squares achieved by adding the term to the current model, defined by the model which includes those terms appearing above the current term given the variance parameters. For example, the test of nitrogen is calculated from the change in sums of squares for the two models mu variety nitrogen and mu variety. No refitting occurs, that is the variance parameters are held constant at the REML estimates obtained from the currently specified fixed model.

The incremental Wald statistics have an asymptotic χ2 distribution, with degrees of freedom (df) given by the number of estimable effects (the number in the DF column). In this example, the incremental F-statistics are numerically the same as the ANOVA F-statistics, and ASReml has calculated the appropriate denominator df for testing fixed effects. This is a simple problem for balanced designs, such as the split plot design, but it is not straightforward to determine the relevant denominator df in unbalanced designs, such as the rat data set described in the next section.

Tables of predicted means are presented for the nitrogen, variety, and variety by nitrogen tables in the .pvs file. The qualifier !SED has been used on the third predict statement and so the matrix of SEDs for the variety by nitrogen table is printed. For the first two predictions, the average SED is calculated from the average variance of differences. Note also that the order of the predictions (e.g. 0.6cwt, 0.4cwt 0.2cwt 0cwt for nitrogen) is simply the order those treatment labels were discovered in the data file.
  Split plot analysis - oat  Variety.Nitrogen                14 Apr 2008 16:15:49
                                                             oats

  Ecode is E for Estimable, * for Not Estimable

  The predictions are obtained by averaging across the hypertable
         calculated from model terms constructed solely from factors
         in the averaging and classify sets.
  Use !AVERAGE to move ignored factors into the averaging set.

  ---- ---- ---- ---- ---- ---- ----   1 ---- ---- ---- ---- ---- ---- ----
  Predicted values of yield
  The averaging set: variety
  The ignored set: blocks wplots

  nitrogen         PredictedValue StandardError Ecode
  0.6cwt                123.3889         7.1747 E
  0.4cwt                114.2222         7.1747 E
  0.2cwt                 98.8889         7.1747 E
  0cwt                   79.3889         7.1747 E
  SED: Overall Standard Error of Difference   4.436

  ---- ---- ---- ---- ---- ---- ----   2 ---- ---- ---- ---- ---- ---- ----
  Predicted values of yield
  The averaging set: nitrogen
  The ignored set: blocks wplots

  variety          PredictedValue  StandardError Ecode
  Marvellous             109.7917         7.7975 E
  Victory                 97.6250         7.7975 E
  GoldenRain             104.5000         7.7975 E
  SED: Overall Standard Error of Difference   7.079

  ---- ---- ---- ---- ---- ---- ----   3 ---- ---- ---- ---- ---- ---- ----
  Predicted values of yield
  The ignored set: blocks wplots

  nitrogen        variety          PredictedValue  StandardError Ecode
  0.6cwt         Marvellous             126.8333         9.1070 E
  0.6cwt         Victory                118.5000         9.1070 E
  0.6cwt         GoldenRain             124.8333         9.1070 E
  0.4cwt         Marvellous             117.1667         9.1070 E
  0.4cwt         Victory                110.8333         9.1070 E
  0.4cwt         GoldenRain             114.6667         9.1070 E
  0.2cwt         Marvellous             108.5000         9.1070 E
  0.2cwt         Victory                 89.6667         9.1070 E
  0.2cwt         GoldenRain              98.5000         9.1070 E
  0cwt           Marvellous              86.6667         9.1070 E
  0cwt           Victory                 71.5000         9.1070 E
  0cwt           GoldenRain              80.0000         9.1070 E

  Predicted values with SED(PV)
     126.833
     118.500        9.71503
     124.833        9.71503        9.71503
     117.167        7.68295        9.71503        9.71503
     110.833        9.71503        7.68295        9.71503        9.71503
     114.667        9.71503        9.71503        7.68295        9.71503
     9.71503
     108.500        7.68295        9.71503        9.71503        7.68295
     9.71503        9.71503
     89.6667        9.71503        7.68295        9.71503        9.71503
     7.68295        9.71503        9.71503
     98.5000        9.71503        9.71503        7.68295        9.71503
     9.71503        7.68295        9.71503        9.71503
     86.6667        7.68295        9.71503        9.71503        7.68295
     9.71503        9.71503        7.68295        9.71503        9.71503
     71.5000        9.71503        7.68295        9.71503        9.71503
     7.68295        9.71503        9.71503        7.68295        9.71503
     9.71503
     80.0000        9.71503        9.71503        7.68295        9.71503
     9.71503        7.68295        9.71503        9.71503        7.68295
     9.71503        9.71503
  SED: Standard Error of Difference: Min   7.6830   Mean   9.1608   Max   9.7150
  • Back

    Return to start