THROWING OUT PLOTS

clearsleepingbagSecurity

Nov 30, 2013 (3 years and 9 months ago)

105 views

THROWING OUT PLOTS

HOW DO YOU KNOW WHEN TO
THROW OUT A PLOT?


MY APPROACH

1.
OBVIOUS POOR STANDS

2.
WHEN UNSURE THEN NOTE PLOTS

3.
EXAMINE DATA AFTER HARVEST

4.
USE DIXON’S TEST FOR OUTLIERS



DIXON’S TEST FOR OUTLIERS

SUBTRACT SUSPECT VALUE WITH NEXT HIGHEST
OR LOWEST AND DIVIDE BY SUSPECT MINUS
LOWEST OR HIGHEST.

GENERAL RULE OF THUMB
-

DIFFERENCE MUST
BE 2 TIMES


FOR SUSPECT LARGE VALUE


X
n



X
n
-
1
/
X
n



X
1


Where
X
n

= Suspect Value


Where X
n
-
1

is the next largest
value


Where X
1

= smallest value

FOR SUSPECT SMALL VALUE


X
n+1

-

X
n

/ X
1

-

X
n



Where
X
n

= suspect small value


Where X
n+1

= next smallest value


Where X
1

= largest value

DIXON’S TABLE OF SIGNIFICANCE

NUMBER

OF VALUES

5%

SIGNIFICANCE LEVEL

3

0.941

4

0.765

5

0.642

6

0.560

EXAMPLE OF USING DIXON’S
TEST FOR LARGE OUTLIER

1.
PLOT VALUES = 20 17 22 23 41

2.
41


23 / 41


17 = .75

3.
PROB. < 0.05

4.
REF. W.J. DIXON, BIOMETRICS 9:89




EXAMPLE OF LOW OUTLIER

1.
PLOT VALUES = 20 7 22 23 19

2.
19
-
7 / 23
-
7= 0.75

3.
PROB. < 0.05

HOW MANY BAD PLOTS
CONSTITUTE A BAD TEST?

MY RULE


IF YOU HAVE TO ADJUST
30% OF THE PLOTS OR MORE THEN
THROW OUT THE ENTIRE TEST OR AT
LEAST THAT REP

IF YOU THROW OUT A PLOT
THEN HOW DO YOU HANDLE IT?

CALCULATE A MISSING PLOT VALUE?
NOT LIKELY.

HOW DOES THIS AFFECT ANY SPATIAL
ANALYSES LIKE NNA OR TREND?

SO HOW DO YOU KNOW WHEN
TO THROW OUT AN ENTIRE
TEST?



1. WHEN THE CV IS HIGH ?

2. HOW HIGH IS TOO HIGH?

3. WHEN YIELDS ARE LOW ?
HOW LOW?


THROW OUT THE TEST WHEN THE
ERROR VARIANCE IS TOO HIGH

1.
NEED HISTORICAL RECORD OF ERROR
VARIANCES FOR EACH CROP

2.
COMPARE SUSPECT TRIAL WITH POOLED
ERROR VARIANCE

3.
IF THE SUSPECT IS 2 TIMES THE POOLED
ERROR VARIANCE THEN THROW THE BUM
OUT

4.
Ref. Bowman and Rawlings. 1995. Agronomy
Journal 87:147
-
151.

WHAT ABOUT WHEN YOU HAVE
LOW YIELDS ?

1.
STUDY IN NC SHOWED THAT DISCARDING
LOW YIELDING TRIALS DID NOT IMPROVE
PREDICTABILITY

2.
TRUE THAT IT IS MORE DIFFICULT TO
SEPARATE MEANS WHEN THEY ARE LOW

3.
TRUE THAT LOW YIELDS CAUSE MORE
QUESTIONS THAN ANSWERS

4.
MY ANSWER
-

DON’T REPORT INDIVIDUAL
LOCATION DATA AND INCLUDE IN ACROSS
LOCATION MEANS

BAD DATA IS WORSE
THAN NO DATA