Agriculture Reference
In-Depth Information
Subsampling
Oftentimes it is desirable or necessary to collect subsamples from
within experimental units. This introduces another source of vari-
ability often called
sampling error
. Such sampling may be desirable
particularly with items that can be easily measured or that are prone
to a great deal of variability. For example, plant height might be bet-
ter represented with several measurements rather than a single plant
within an experimental unit, while measuring every plant in the
experimental unit may be too time consuming or costly.
Open the dataset Watermelon Subsampling.dta. This is a dataset
from a variety trial where two fruits from each experimental unit were
measured for length, width, rind thickness, and percent soluble solids
(sugar content). In addition, there is a variable representing the ratio
of the length to width (
lwratio
). Enter the following command and
see the results:
anova
lwratio rep trt
/
rep
#
trt
Number of obs = 168 R-squared = 0.8887
Root MSE = .130343 Adj R-squared = 0.7788
Source | Partial SS df MS F Prob > F
-----------+----------------------------------------------------
Model | 11.3971347 83 .137314876 8.08 0.0000
|
rep | .203931699 3 .067977233 1.04 0.3798
trt | 7.28573298 20 .364286649 5.59 0.0000
rep#trt | 3.90747006 60 .065124501
-----------+----------------------------------------------------
|
Residual | 1.42711017 84 .016989407
-----------+----------------------------------------------------
Total | 12.8242449 167 .076791886
The experimental error term (denominator) to calculate the
F-test, in this case, is the replication by treatment interaction
(
rep#trt
). In an RCBD without subsampling, the experimental
error term would simply be the residual. Table 5.5 shows the cor-
rect terms to use in calculating a CRD, RCBD, and a split-plot
design with subsampling. The important thing to note is that sam-
pling error has been accounted for and that the appropriate error
term is used.
Search WWH ::
Custom Search