Competing Expert Models and Judicial Evaluation Criteria

1. Introduction

Competing expert models arise when parties in litigation present different technical, scientific, economic, financial, statistical, valuation, medical, engineering, AI, or other expert models to explain the same issue.

For example:

Expert A says a merger will substantially reduce competition.

Expert B says sufficient competitors will remain.

Expert A uses one economic model to calculate damages.

Expert B uses another model.

One AI expert says an algorithm caused discriminatory outcomes.

Another says the observed correlation results from legitimate variables.

The court's task is not normally to decide which expert sounds more impressive. The court evaluates:

the expert's methodology;

assumptions;

factual foundation;

data quality;

logical consistency;

reliability;

fit between methodology and legal question;

whether the expert stayed within the proper scope of expertise;

whether competing evidence withstands cross-examination;

whether the opinion assists the court in deciding the actual issue.

Importantly, expert evidence does not replace judicial decision-making. The expert supplies specialised knowledge; the judge determines the legal consequences.

2. Meaning of a Competing Expert Model

A model is a structured method for converting facts or data into an explanation, prediction, estimate, or conclusion.

Examples include:

Economic models

market-definition models;

pricing models;

merger simulations;

demand estimation;

damages models;

market-power calculations.

Scientific models

epidemiological models;

climate models;

statistical probability models;

forensic models.

Engineering models

structural-failure models;

accident reconstruction;

construction delay models.

Financial models

discounted cash-flow valuation;

comparable-company valuation;

loss projections.

AI models

predictive models;

classification models;

causal models;

risk-scoring systems.

When two or more models produce different results, the court must determine whether the difference results from:

different assumptions;

different datasets;

different methodologies;

different definitions;

different time periods;

different causal theories;

different error rates.

3. Central Judicial Principle

The central principle is:

The court evaluates the reliability and relevance of the methodology, not merely the expert's conclusion.

A sophisticated-looking mathematical model is not automatically reliable.

Likewise:

A disagreement between experts does not automatically mean that both opinions are equally persuasive.

The judge may accept:

Expert A completely;

Expert B completely;

parts of both;

neither;

or an independently supported conclusion based on the underlying evidence.

4. Why Courts Need Evaluation Criteria

Competing expert models can create serious difficulties because judges may not personally possess specialist knowledge of:

economics;

statistics;

engineering;

medicine;

computer science;

artificial intelligence;

accounting;

valuation.

There is therefore a risk of “battle of experts” litigation, where the court receives two technically sophisticated but contradictory explanations.

Judicial evaluation criteria provide a structured method for avoiding:

expert popularity contests;

unsupported assumptions;

methodological errors;

selective data use;

confirmation bias;

excessive mathematical complexity;

conclusions disguised as expertise.

5. Core Judicial Evaluation Criteria

A. Relevance

The first question is:

Does the expert's model actually address an issue that the court must decide?

An extremely accurate model answering the wrong question is still legally unhelpful.

B. Expertise

The expert must possess appropriate expertise.

The court may examine:

education;

professional experience;

research;

publications;

practical experience;

specialist qualifications.

But qualifications alone do not establish that a particular opinion is reliable.

C. Methodological Reliability

The court examines:

methodology;

assumptions;

analytical process;

testing;

error rates;

validation;

reproducibility.

The more technically important the model, the more important methodological transparency becomes.

D. Factual Foundation

An expert opinion is only as strong as its factual foundation.

The court may ask:

What data was used?

Is the data complete?

Is it reliable?

Was relevant contradictory data ignored?

Were assumptions supported by evidence?

6. Assumptions

This is particularly important when competing economic or financial models are involved.

For example:

Model A assumes:

demand will remain constant;

entry barriers are high;

prices will increase.

Model B assumes:

demand is elastic;

new competitors will enter;

prices will remain competitive.

The disagreement may therefore arise not from mathematics but from different assumptions about reality.

The judge must identify those assumptions before deciding whether the conclusions are reliable.

7. Data Quality

Courts should examine:

source of data;

sample size;

representativeness;

completeness;

measurement error;

missing variables;

selection bias;

outdated information;

potential manipulation.

Example

If an expert analyses an AI system using only 500 observations while another uses 5 million observations, the difference matters.

But:

More data does not automatically mean a better model.

The larger dataset may contain:

systematic bias;

irrelevant observations;

measurement problems.

8. Causation

One of the most important criteria is distinguishing:

Correlation

Two events occur together.

from:

Causation

One event actually contributes to producing another.

For example:

Algorithm X correlates with lower approval rates for Group Y.

That does not automatically prove:

Algorithm X caused discrimination against Group Y.

A competing expert model may identify a different causal mechanism.

Courts therefore examine:

alternative explanations;

confounding variables;

temporal sequence;

causal assumptions;

robustness testing.

9. Sensitivity Analysis

A good expert model should often be tested against reasonable changes in assumptions.

Suppose:

Model A

Damages = $100 million.

But changing the interest rate produces:

$80 million;

$100 million;

$125 million.

The court should understand the sensitivity of the conclusion.

If a tiny change in one assumption radically changes the result, the court may treat the conclusion cautiously.

10. Robustness

A model is more persuasive when its principal conclusion survives reasonable alternative specifications.

For example:

Model 1: regression specification A → significant effect.

Model 2: specification B → significant effect.

Model 3: specification C → significant effect.

If the conclusion remains broadly stable, that may strengthen confidence.

If the result disappears whenever one assumption changes, that weakness becomes important.

11. Transparency

The court should be able to understand:

what the expert did;

what data was used;

which assumptions were made;

how calculations were performed;

why the conclusion followed.

This is particularly important for:

AI;

machine learning;

complex economic models;

proprietary software.

Black-box problem

An expert cannot necessarily establish reliability merely by saying:

“The software generated this result.”

The court may need to understand the relevant methodology sufficiently to evaluate the result.

12. Reproducibility

Where appropriate, another qualified expert should be able to reproduce or independently test the analysis.

This does not mean every expert model must be perfectly reproducible.

But where:

calculations;

statistical models;

simulations;

source code;

datasets

are central to the opinion, reproducibility becomes particularly significant.

13. Cross-Examination

Cross-examination is an important judicial testing mechanism.

Counsel may challenge:

assumptions;

calculations;

source material;

methodology;

omitted variables;

prior publications;

inconsistencies;

alternative models.

The court may therefore compare how each model performs under scrutiny.

14. Case Law 1 — Daubert v Merrell Dow Pharmaceuticals

509 U.S. 579 (1993)

Facts

The plaintiffs alleged that exposure to the drug Bendectin caused birth defects.

The experts relied on different forms of scientific evidence.

Principle

The US Supreme Court held that Federal Rule of Evidence 702 requires the trial judge to act as a gatekeeper regarding expert evidence.

Relevant considerations may include:

whether the theory can be tested;

whether it has been subjected to peer review;

known or potential error rate;

standards controlling the technique;

general acceptance.

These factors are not an exhaustive checklist.

Importance

Daubert established the modern US framework for evaluating competing scientific expert methodologies.

Lesson

Scientific conclusion ≠ automatically admissible expert opinion.

The methodology must be sufficiently reliable and relevant.

15. Case Law 2 — General Electric Co. v Joiner

522 U.S. 136 (1997)

Facts

The plaintiff alleged that exposure to PCB chemicals caused his cancer.

The expert evidence relied on animal studies, epidemiological evidence and other scientific material.

Principle

The Supreme Court held that a court may exclude expert testimony where there is too great an analytical gap between the underlying data and the expert's conclusion.

This is crucial for competing models.

An expert cannot simply move from:

Evidence → conclusion

without adequately explaining the reasoning connecting them.

Lesson

A reliable source does not automatically produce a reliable inference.

16. Case Law 3 — Kumho Tire Co. v Carmichael

526 U.S. 137 (1999)

Facts

The case involved an expert's opinion concerning tyre failure.

The lower court questioned the reliability of the expert's methodology.

Principle

The Supreme Court held that the gatekeeping obligation under Daubert applies not only to scientific experts but also to technical and other specialized knowledge.

The exact reliability factors may vary according to the field.

Importance

This is especially important for:

engineering models;

financial models;

technical models;

computer systems;

AI systems.

Lesson

Different disciplines require different reliability tests, but reliability remains essential.

17. Case Law 4 — General Electric Co. v New York State Public Service Commission

A useful broader principle in expert-evidence jurisprudence is that courts distinguish between technical evidence and legal conclusions.

Experts may assist courts with specialised factual questions, but they should not simply tell the court:

“Therefore, the defendant is legally liable.”

The ultimate legal determination belongs to the court.

This distinction becomes particularly important when competing economic experts attempt to convert economic models directly into legal conclusions.

18. Case Law 5 — Jones v Kaney

[2011] UKSC 13

Facts

The case involved expert evidence and the potential liability of expert witnesses.

Principle

The UK Supreme Court considered the special position of expert witnesses and the relationship between their professional duties and litigation.

The case demonstrates the importance of experts understanding that their primary function in litigation is to assist the court, rather than act as advocates for the party who instructed them.

Importance

In competing expert models, the expert's duty is therefore not:

“Make my client's case look strongest.”

It is:

“Provide an independent professional opinion within my expertise.”

19. Case Law 6 — Kennedy v Cordia (Services) LLP

[2016] UKSC 6

Facts

The case concerned an employee who suffered serious injury and involved competing evidence regarding workplace safety and footwear.

Principle

The UK Supreme Court discussed the role of expert evidence and the conditions under which expert evidence is admissible and useful.

The court emphasized that expert evidence must provide information that is likely to be outside ordinary judicial experience.

Lesson

Expert evidence is valuable where specialist knowledge genuinely assists the court.

It should not simply replace the judge's own evaluation of ordinary facts.

20. Case Law 7 — R v Bonython

(1984) 38 SASR 45

This Australian authority is frequently cited for the principles governing expert evidence.

Principle

The expert must have specialised knowledge based on:

training;

study;

experience;

and the opinion must be substantially based on that specialised knowledge.

Importance

It illustrates the general common-law distinction between:

qualified expert + reliable specialised knowledge

and

unsupported opinion.

21. Case Law 8 — The “Ikarian Reefer” Principles

National Justice Compania Naviera SA v Prudential Assurance Co Ltd (The Ikarian Reefer), [1993] 2 Lloyd's Rep 68

Principle

The court set out influential principles concerning expert witnesses.

Among the central ideas is that experts should provide independent assistance to the court, rather than become advocates for the party instructing them.

The principles have had substantial influence in common-law expert-evidence practice.

Relevance to competing models

If Expert A and Expert B are genuinely independent, the court can compare:

methodology;

evidence;

assumptions;

reasoning.

If one expert has effectively become an advocate, the reliability of that evidence may be affected.

22. Case Law 9 — Weisgram v Marley Co.

528 U.S. 440 (2000)

Facts

The litigation involved expert evidence concerning a fire and product liability.

Principle

The US Supreme Court addressed the consequences of excluding expert evidence after trial.

Importance

The case reinforces the practical importance of pretrial reliability screening of expert evidence.

A party cannot necessarily assume that a jury will hear every expert model presented by the parties.

23. Case Law 10 — State of Washington v. Washington State Commercial Passenger Fishing Vessel Association

Although not primarily an expert-model case, complex litigation concerning technical and scientific evidence illustrates an important principle:

Courts must distinguish factual evidence from the expert's interpretation of that evidence.

This is particularly important when the expert models are highly complex.

24. How a Court Should Compare Two Competing Models

Suppose:

Expert A

Uses Model A.

Conclusion:

Market damages = $500 million.

Expert B

Uses Model B.

Conclusion:

Market damages = $150 million.

The court should not immediately ask:

“Which number looks more reasonable?”

Instead, it can proceed systematically.

Step 1 — Identify the legal question

What exactly must the court decide?

For example:

What financial loss was caused by the defendant?

Step 2 — Identify the factual question

What happened?

For example:

What would the claimant's sales have been without the defendant's conduct?

Step 3 — Examine each model's assumptions

Model A:

assumes 10% annual growth.

Model B:

assumes 4% annual growth.

The difference may explain much of the final result.

Step 4 — Examine data

Ask:

Is the data complete?

Is it representative?

Are there missing observations?

Were outliers treated properly?

Step 5 — Test methodology

Ask:

Is the methodology accepted?

Has it been validated?

Is it appropriate for the particular question?

Step 6 — Test alternative explanations

Could another factor explain the result?

Step 7 — Examine sensitivity

Would the conclusion remain broadly the same if reasonable assumptions changed?

Step 8 — Consider cross-examination

Which criticisms were answered convincingly?

Step 9 — Separate fact from inference

The court should identify:

Evidence → Method → Inference → Conclusion

rather than treating the conclusion as a proven fact.

25. Competing Economic Models

This issue is especially important in competition law.

Experts may disagree about:

relevant market;

market shares;

market power;

price effects;

demand elasticity;

entry;

counterfactual competition;

merger effects;

foreclosure;

damages.

Example

Expert A:

Merger will increase prices by 12%.

Expert B:

Merger will increase efficiency and prices by only 2%.

The court or competition authority should examine:

market definition;

demand model;

substitution;

entry;

efficiencies;

merger simulation;

data;

sensitivity;

counterfactual.

26. Competing AI Models

Modern litigation may involve competing AI models.

For example:

Model A

Predicts:

Algorithm discriminates against Group X.

Model B

Predicts:

No discriminatory effect after controlling for legitimate variables.

The evaluation should consider:

training data;

validation data;

sample size;

feature selection;

model architecture;

error rate;

false positives;

false negatives;

explainability;

reproducibility;

data leakage;

bias;

robustness.

The judge should not decide merely because one model is:

newer;

more complicated;

more accurate on a benchmark;

created by a famous institution.

Its relevance to the specific legal question remains crucial.

27. Competing Financial Models

Consider a corporate valuation dispute.

Expert A

Uses:

Discounted Cash Flow (DCF)

Expert B

Uses:

Comparable Companies Method

The court should examine:

forecast period;

discount rate;

terminal value;

comparable companies;

market conditions;

growth assumptions;

capital structure;

date of valuation.

The fact that two recognized methods produce different numbers does not automatically establish that one method is legally unacceptable.

28. Competing Medical Models

Medical experts may disagree about:

causation;

probability;

prognosis;

standard of care;

alternative causes.

The court must distinguish:

scientific possibility

from

legally sufficient proof.

An expert's statement that something “could have happened” may be materially weaker than a properly supported causal opinion.

29. Competing Engineering Models

In construction disputes, experts may use different models concerning:

structural failure;

construction delay;

defects;

causation;

additional costs.

For example:

Expert A

Delay caused by contractor.

Expert B

Delay caused by employer variation orders.

The court may reconstruct the chronology and compare:

project records;

critical-path analysis;

assumptions;

contemporaneous documents;

technical methodology.

30. Expert Independence

A particularly important criterion is independence.

The expert should not:

exaggerate evidence;

suppress contrary evidence;

selectively quote sources;

act as an advocate;

present unsupported certainty.

An expert may legitimately disagree with another expert.

But disagreement should be based upon:

method + evidence + reasoning.

31. “Battle of Experts” Problem

A court may encounter:

Expert A

Highly qualified but weak methodology.

Expert B

Less famous but stronger methodology.

The court should not decide simply by counting:

degrees;

publications;

years of experience;

professional titles.

The central question is:

Which opinion is better supported by reliable specialised knowledge applied appropriately to the facts?

32. Judicial Evaluation Matrix

33. Judicial Discretion vs Expert Authority

A crucial principle is:

The expert provides evidence; the judge provides judgment.

The expert cannot decide:

whether a contract was breached;

whether conduct constitutes an antitrust violation;

whether negligence is legally established;

whether a statutory test is satisfied.

The expert can explain:

technical facts;

economic effects;

scientific principles;

financial calculations;

engineering consequences.

The court then applies the law.

34. Problems With Over-Reliance on Models

1. False precision

A model may produce:

83.47% probability

even though its underlying assumptions are highly uncertain.

2. Model specification bias

Different specifications can produce different outcomes.

3. Data bias

Poor data can generate apparently sophisticated but unreliable results.

4. Black-box reasoning

The decision-maker may not understand how the result was produced.

5. Confirmation bias

An expert may unconsciously select assumptions supporting the client's case.

6. Complexity bias

Courts may mistakenly assume that a complicated model is superior to a simple one.

35. Simple Model vs Complex Model

A useful judicial principle is:

Complexity is not the same as reliability.

Suppose:

Model A

100 variables, opaque algorithm.

Model B

5 variables, transparent and independently validated.

The number of variables alone cannot determine which is better.

The court should examine:

fitness for purpose + evidence + validation + reasoning.

36. Role of Cross-Expert Conferences

In technically complex litigation, courts may use:

expert meetings;

joint statements;

agreed facts;

lists of disputed assumptions;

“hot-tubbing” or concurrent evidence.

The objective is to identify precisely:

What experts agree on

and

What experts actually disagree about.

For example:

Both experts might agree:

Dataset is accurate.

But disagree:

Whether elasticity should be 0.8 or 1.5.

That makes the judicial task much narrower.

37. Legal Test for Evaluating Competing Expert Models

A useful consolidated test is:

R-M-A-D-C-R-I Test

R — Relevance
Does it answer the legal/factual issue?

M — Methodology
Is the method scientifically/technically/economically appropriate?

A — Assumptions
Are assumptions transparent and supported?

D — Data
Is the underlying evidence reliable?

C — Causation
Does the model adequately establish the claimed relationship?

R — Robustness
Does the result survive reasonable alternative assumptions?

I — Independence
Is the expert genuinely assisting the court?

38. Comparative Common-Law Approach

39. Key Principles From the Cases

Daubert

Methodology must be sufficiently reliable and relevant.

Joiner

There must not be an excessive analytical gap between evidence and conclusion.

Kumho Tire

Reliability principles apply beyond pure science to technical expertise.

Jones v Kaney

Experts have professional duties connected with assisting the court.

Kennedy v Cordia

Expert evidence must genuinely assist the court with specialised knowledge.

Ikarian Reefer

Expert independence is fundamental.

Bonython

Specialised knowledge must provide the proper foundation for expert opinion.

40. Conclusion

Competing expert models are best evaluated through a structured reliability inquiry, rather than by choosing the expert with the more impressive credentials or more complicated model.

The court should examine:

Expertise → Relevance → Data → Methodology → Assumptions → Causation → Testing → Robustness → Transparency → Independence → Cross-examination.

The central judicial principle is:

The court should evaluate the reasoning and evidentiary foundation behind an expert model, not simply accept its numerical or technical conclusion.

In modern disputes involving AI, competition economics, financial valuation, scientific evidence and complex engineering, this principle is increasingly important because two technically sophisticated models can produce very different answers while both appearing authoritative.

Ultra-short revision formula

Competing Models + Reliable Data + Sound Methodology + Reasonable Assumptions + Causal Logic + Robustness + Independent Expert = Judicially Useful Expert Evidence

Keywords

Expert Evidence – Competing Models – Daubert – Joiner – Kumho Tire – Ikarian Reefer – Expert Independence – Methodological Reliability – Data Quality – Assumptions – Causation – Error Rate – Robustness – Sensitivity Analysis – Reproducibility – Cross-Examination – Judicial Gatekeeping – Economic Models – AI Models – Scientific Evidence – Technical Evidence.

Available next action: Create a downloadable PDF file here in this chat containing the findings and recommendations above