Comparison of stickiness results from different instruments, observations based on an international round-trial

Centre de coopération internationale en recherche agronomique pour le développement

ITMF, ICCTM

Stickiness Working group

Comparison of Stickiness results

from different instruments,

observations based on an

International Round-trial

Bremen, April 1, 2008

Gozé E.,

Frydrych R., Gourlot J.-P., Lassus S.

Cotton Technology Laboratory, Montpellier, France

Outline of presentation

9

Reasons for organizing a round-test

9

Experimental design and its

implementation

Outline of presentation

9

Reasons for organizing a round-test

9

Experimental design and its

implementation

Reasons for organizing a round-test

• Mini-card

• Fibre Contamination Tester (FCT) and

Fibre Quality Tester (FQT),

• Stickiness Cotton Thermodetector (SCT),

• High Speed Stickiness Detector (H2SD)‏

• …

For stickiness measurement, several types of

instruments exist:

Reasons for organizing a round-test

• to quantify the accuracy of the results

obtained from each type of instrument for

a given cotton

• to check the relationship between results

obtained from various types of

instruments.

• Lack of accuracy is usually splitted into

two components : (ISO 5725-1)

– Bias

(<> trueness)

– Variability

(<> precision)

The bias is the mean departure from a

true value determined with a reference

instrument.

The variability is the departure between

measurements made with the same

instrument type.

Bias

• Here the reference instruments are

minicard and SCT…

• … but the other intruments are not aimed

at measuring the same quantity.

• Even the two reference instruments differ:

the minicard output is qualitative, the SCT

is quantitative, and so are the other

Variability

• within samples

• within laboratories

• between laboratories

The variability of measurements has several

components :

repeatability

reproducibility

Outline

9

Reasons for organizing a round-test

9

Experimental design and its

implementation

Experimental design

• homogenizing the material has been shown (ITMF,

2002) to reduce the variance of this distribution :

separate round tests should be carried out for

mixed and raw cotton

Basics assumptions for preparing the round

test:

Experimental design

We then included various cottons

- covering a range of stickiness

with 6 cottons min, 10 recommended

- contrasting between two sample preparations

Raw cottons

Mixed cottons

-With at least 8 laboratories

The recommended number of laboratories has not been achieved

and these results should be considered as preliminary before a

larger round test can be organized.

Otherwise, the round test design is not different from the one

suitable for gaussian measurements : randomized in blocks within

laboratories, with blind measurements.

Range of stickiness

R11

45

11

R12

12

12

R10

30

10

R9

10

9

R8

M8 (2 blocks)‏

20

8

R7

M7 (2 blocks)‏

17

7

R6

M6 (2 blocks)‏

30

6

R5

M5 (2 blocks)‏

50

5

R4

M4 (2 blocks)‏

15

4

R3

M3 (2 blocks)‏

80

3

R2

M2 (2 blocks)‏

0

2

R1

M1 (2 blocks)‏

50

1

Coton Raw ( R)

Mixed Cotton (M)‏

Stickiness level

(H2SD)‏

Cotton

Organization of the round-test

Fiber bank

(Cotton, Stickiness data)‏

Cottons selection (1 to 12)‏

Cotton 1 to 12

Manual homogenization

Sampling

Raw cotton ( R)‏

Mixed cotton (M)‏

Card without flats

+ large drum

Samples

preparation

(R1)‏

Samples

preparation

(M1)‏

Organization of the round-test

Cotton 1

Envelopes R1

Envelopes M1

Cotton 2

Envelopes R2

Envelopes M2

Cotton n

Envelopes Rn

Envelopes Mn

Box R(A)‏

Box M(A)‏

Labo

aa

Box R(B)‏

Box M(B)‏

Labo

bb

Cotton …

Round-test

2008

Envelopes are randomized in

complete blocks within each box

Plan of presentation

9

Reasons for organizing a round-test

9

Organisation of the Round Test

Number of copies of each instrument

0

4 (5?)‏

3

3

2

Achieved

0

4 (5?)‏

3

3

1 (2?)‏

Achieved

Planned

Planned

1

1

HPLC

6

5

H2SD

4

4

FCT/FQT

5

5

SCT

3

3

Mini-card

Mixed

Raw

Instrument

L o g N ( 1 + v a r i a n c e ) v s L o g N ( 1 + m e a n ) e t a t = m i x e d A p p a r e i l = F C T _ F Q T l v 0 1 2 3 4 5 6 7 8 l m 0 1 2 3 4 5 6 7 8

L o g N ( 1 + v a r i a n c e ) v s L o g N ( 1 + m e a n ) e t a t = r a w A p p a r e i l = F C T _ F Q T l v 0 1 2 3 4 5 6 7 8 l m 0 1 2 3 4 5 6 7 8

L o g N ( 1 + v a r i a n c e ) v s L o g N ( 1 + m e a n ) e t a t = m i x e d A p p a r e i l = H 2 S D l v 0 1 2 3 4 5 6 7 8 l m 0 1 2 3 4 5 6 7 8

L o g N ( 1 + v a r i a n c e ) v s L o g N ( 1 + m e a n ) e t a t = r a w A p p a r e i l = H 2 S D l v 0 1 2 3 4 5 6 7 8 l m 0 1 2 3 4 5 6 7 8

L o g N ( 1 + v a r i a n c e ) v s L o g N ( 1 + m e a n ) e t a t = m i x e d A p p a r e i l = S C T l v 0 1 2 3 4 5 6 7 8 l m 0 1 2 3 4 5 6 7 8

L o g N ( 1 + v a r i a n c e ) v s L o g N ( 1 + m e a n ) e t a t = r a w A p p a r e i l = S C T l v 0 1 2 3 4 5 6 7 8 l m 0 1 2 3 4 5 6 7 8

Cotton i

Cotton i Lab j

Cotton x lab interaction

Block within lab effect

Neperian log scale

• E[Yijk] = Mi . Bj . (MB)ij . Cjk

Conditional distribution

• E[Yijk] = Mi . Bj . (MB)ij . Cjk

• Log(E[Yijk]) = m + a

i

+ b

j

+ (ab)

ij

+ c

jk

• Y | E[Y] = given its expection, Y follows

a negative binomial of parameter k

with overdispersion ø

• Results were analyzed with the generalized linear

model procedure of Sas (proc genmod)

• One device had incoherent results, and broke down

shortly after the test : its data was discarded.

• Otherwise, the inspection of residuals did not show

i n s t r _ p r e p F C T - F Q T m i x e d F C T - F Q T r a w H 2 S D m i x e d H 2 S D r a w S C T m i x e d S C T r a w o v e r d i s p 0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 m e a n 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 1 1 0 1 2 0 1 3 0 1 4 0 1 5 0

i n s t r _ p r e p F C T - F Q T m i x e d F C T - F Q T r a w H 2 S D m i x e d _ ( n H 2 S D r a w _ ( n e g S C T m i x e d S C T r a w s t d e v 0 1 0 2 0 3 0 4 0 5 0 6 0 m e a n 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 1 1 0 1 2 0 1 3 0 1 4 0 1 5 0

i n s t r _ p r e p F C T - F Q T m i x e d F C T - F Q T r a w H 2 S D m i x e d _ ( n H 2 S D r a w _ ( n e g S C T m i x e d S C T r a w c v 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 m e a n 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0 1 1 0 1 2 0 1 3 0 1 4 0 1 5 0

Blocks and lab effects

Yes/No

+192%

+230%

+63%

+28%

SCT

Yes/No

+300%

+310%

+85%

+41%

FQT

No/No

+85%

+43%

+33%

+25%

H2SD

Instrumen

t

Raw

Mixed

Raw

Mixed

Preparation

Labxcot

inter

Max difference between

labs means

Max difference between

Yes/No

+192%

+230%

+63%

+28%

SCT

Yes/No

+300%

+310%

+85%

+41%

FQT

No

/No

+85%

+43%

+33%

+25%

H2SD

Instrumen

t

Raw

Mixed

Raw

Mixed

Preparation

Labxcot

inter.

Max difference between

labs means

Max difference between

blocks means

Yes/No

+192%

+230%

+63%

+28%

SCT

Yes

/No

+300%

+310%

+85%

+41%

FQT

No/No

+85%

+43%

+33%

+25%

H2SD

Instrumen

t

Raw

Mixed

Raw

Mixed

Preparation

Labxcot

inter.

Max difference between

labs means

Max difference between

blocks means

SCT 3.4848 118.9697 H2SD 5.7045 63.2326 FCT_FQT 4.2424 148.7576 Car d 1.7368 7.0000

Summary

• These are preliminary results, too few labs per instrument

• Single CV calculation of analysis of variance are not

appropriate, but a generalized linear model gives sensible

results :

– Within lab precision is not the same for all the

instruments