PROBLEMS OF INTERCALIBRATIONS

1
PROBLEMS OF INTERCALIBRATIONS

• Kovanic P. and Ocelka T.
• Institute of Public Health, Ostrava, Czech
  Republic

2
EXPERIENCE FROM THE GEOLOGY

• International interlaboratory calibration exists
• in geology for decades
• Participants National Geological Institutions
  and specialized private firms
• Goals
• To establish RVs (Recommended Values)
• of CRMs (Certified Reference Materials)
• To certify methods and laboratories

3
THE PROCESS

• Selection of the target material
• Selection of the country and location with the
• most representative type of the material
• Accumulation and grinding of the material
• Preparing standard portions/samples
• Distributing to laboratories worldwide
• (example 128 laboratories of 29 countries)
• Centralization and treatment of results

4
PROBLEMS OF THE DATA TREATMENT

• Different laboratories apply different methods
• Elementary statistical methods are required by
  norms and applied to handle data
• These methods can be inadequate
• relying on a priori (subjective) data models,
• requiring large data samples,
• not being robust,
• not using censored data properly,
• not yielding maximum of data information.

5
EXAMPLE Cobalt IN GRANIT

• Practical aspect storages of nuclear wastes
• The selected country in 90-th Czechoslovakia
• Realization The Czech Geological Survey, Prague
• Data from 40 laboratories
• Results published
• Paukert T., Rubeška I., Kovanic P.,
• ANALYST, Febr. 1993, Vol.118, 145-148
• Surprise 3-modal probability density

6
(No Transcript)
7
DERIVED QUESTIONS

• Which standard distribution function would
• fit the data?
• What would be the reference value?
• Is the range of possible data bounded?
• If yes, then what values have the bounds?
• Are some data outliers?
• Way out Use distribution functions determined
• by data instead of point estimates to get
  all
• information. Let data speak for themselves!

8
(No Transcript)
9
(No Transcript)
10
IMPORTANCE OF ESTIMATED DATA BOUNDS

• Data bounds LB and UB, D is a data item
• Statements on probability/expectation P of
• PD lt LB 0 and PUB lt D 0
• hold almost surely
• Bounds LB and UB can be the most important
• (even vital) result of an investigation

11
(No Transcript)
12
(No Transcript)
13
SIMPLE QUESTION IS x AN OUTLIER?

• Standard approach to a given data sample D
• Estimate mean F of D and standard deviation s
• Choose significance ?1, ?2 or ?3
• Test X is an outlier if
• a) abs(X - F) gt s (significance ?1)
• b) abs(X - F) gt 2s (significance ?2)
• c) abs(X - F) gt 3s (significance ?3)
• SUBJECTIVITY 1) assumption of normality of D
• 2) choice of the
  significance
• ALTERNATIVE Mathematical Gnostics

14
(No Transcript)
15
(No Transcript)
16
MEASURING AND RISKS

• To decide, we need statements on state of things.
• Risks of deciding depend on verity of statements.
  
• Quantified statements are derived by treatment of
  data resulting from measurements.
• Risks are measured by probabilities.
• Data treatment must evaluate not only true data
  values masked by uncertainty, but also the risks
  involved by uncertainty.
• Probability distributions are a must of treatment

17
(No Transcript)
18
RATING OF METHODS
Feature Bentos Dreiss. Plants SPMD
Ext.consistency 4 3 1 2
Int.consistency 4 3 2 1
Informativeness 1 3 4 2
Precission 3 1 4 2
Homogeneity 2 4 3 1
Rel.sensitivity 3 1 2 1
Mean rating 2.8 2.5 2.7 1.5
19
(No Transcript)
20
AVAILABILITY OF THE ALTERNATIVE

• Tools of mathematical gnostics are being
  developed in the frame-work of EU grants
• More info
• http//www.2-fun.org/download/newsletter4_en
• Goal Package of gnostic software for the GNU
  mathematical and statistical environment freely
• available on http//www.R-project.org

21
CONCLUSIONS

• Intercalibration is a complex task
• Traditional (statistical) tools are not always
  adequate
• Non-conservative alternative exists worth a try

22
Funding
European Commission Sixth Framework Program,
Priority 6 (Global change and ecosystems),
project 2-FUN (contract036976)