Vous êtes sur la page 1sur 17

Measurement and Control of Experimental Error

Bulls eye represents the true value of the parameter you wish to estimate Accuracy = without bias average is on the bulls-eye achieved through randomization Precision = repeatability measurements are close together achieved through replication

Both accuracy and precision are needed!

Measurement and Control of Experimental Error


Randomization
To eliminate bias To ensure independence among observations. This is required for valid significance tests and interval estimates.

Replication
The repetition of a treatment in an

experiment

A B D A
C D B C B A D C

Measurement and Control of Experimental Error


Replication Each treatment is applied independently to two or more experimental units Variation among plots treated alike can be measured Increases precision - as n increases, error decreases
Standard error of a mean Sample variance Number of replications

broadens the base for making inferences - smaller differences can be detected

Effect of number of replicates


Effect of replication on variance
8.0 7.5 7.0 6.5 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0

Variance of the mean

10

15

20

25

30

35

40

45

50

number of replicates

What determines the number of Replications?


Pattern and magnitude of variability in the soils Size of the difference to be detected Required significance level Number of treatments Amount of resources that can be devoted to the

experiment Limitations in cost, labor, time, and so on

Control of Experimental Error


Select appropriate experimental units Increase the size of the experiment to gain more

degrees of freedom (more replicates or more treatments)


caution - s2 will increase as more heterogeneous material is used - may be self-defeating

Select appropriate treatments factorial combinations result in hidden replications and therefore will increase n Blocking

Control of Experimental Error


Blocking A block is a group of homogeneous experimental units Maximize the variation among blocks in order to minimize the variation within blocks Reasons for blocking To remove block to block variation from the experimental error (increase precision) Treatment comparisons are more uniform Increase the information by allowing the researcher to sample a wider range of conditions

Blocking
At least one replication is grouped in a

homogeneous area

A B D A
C D B C B A D C
Just replication

A B D C
C D B A B A D C
Blocking

Control of Experimental Error


Criteria for blocking Proximity or known patterns of variation in the field Physical characteristics (height, maturity) Time (planting, harvesting) Management of experimental tasks (individuals collecting data, etc.)

The Field Plot


The experimental unit: the vehicle for evaluating

the response of the material to the treatment Shapes


Rectangular is most common - run the long dimension parallel to any gradient Fan-shaped may be useful when studying densities Shape may be determined by the machinery or irrigation

Plot Shape and Orientation


Long narrow plots are preferred

usually more economical for field operations all plots exposed to the same conditions If there is a gradient - the longest plot dimension should be in the direction of the greatest variability

Border Effects
Plants along the edges of plots often perform differently

than those in the center of the plot Border rows on the edge of a field or end of a plot have an advantage in competition Plants on the perimeter of the plot can be influenced by plant height or competition from adjacent plots Machinery can drag the effects of one treatment into the next plot Fertilizer or irrigation can move from one plot to the next Impact of border effect is greater with very small plots

Minimizing Border Effects


Leave alleys between plots to minimize drag Remove plot edges and measure yield only on

center portion Plant border plots surrounding the experiment

Rounding and Reporting Numbers


To reduce measurement error:
Standardize the way that you collect data

and try to be as consistent as possible Actual measurements are better than subjective readings When collecting data - examine out-of-line figures immediately and recheck

Rounding in ANOVA
In doing an ANOVA, it is best to carry the

full number of figures obtained from the uncorrected sum of squares


If, for example, the original data contain one decimal, the sum of squares will contain two places 2.2 * 2.2 = 4.84
n

Do not round closer than this until reporting final results

Rounding
General rule: When rounding numbers the digit to be

retained is rounded upward if the digit to be dropped is greater than 5 or is 5 followed by a digit greater than zero. 21.549 rounded 21.5

21.552

rounded

21.6

Optional rule: If the amount following the 5 is zero, the

digit to be rounded is rounded upward if odd, or left as is if it is even. 21.550 21.450 rounded rounded 21.6 21.4