1 of 15

Science, Ontology, and Causal Analysis

Based in part on work done with:

Anandi Hira of Tecolote Research (formerly) and the Software Engineering Institute.

Mike Konrad of the Software Engineering Institute (SEI) at Carnegie Mellon.

Anandi, Mike, and Winsor Brown contributed to revisions.

​

​

​

1

11/9

Jim Alstad

alstad@acm.org

10 November 2022

​

​

​

University of Southern California

Center for Systems and Software Engineering

2 of 15

Agenda

  • Some basics about causal graphs
  • A scientific inquiry into some phenomena
  • Causal analysis vs the scientific method
  • Conclusion

​

​

2

11/9

University of Southern California

Center for Systems and Software Engineering

3 of 15

Notation for Causal Graphs (1/2)

3

11/9

Circle: A phenomenon under study.� A phenomenon must be

quantifiable:

Either numeric,

or a choice among a

finite set of possibilities; and

consistently measured.

Solid circle: Observable phenomenon.

Dashed circle: Phenomenon is not observable.

University of Southern California

Center for Systems and Software Engineering

4 of 15

Notation for Causal Graphs (2/2)

4

11/9

A

B

Phenomenon A is an immediate cause of phenomenon B (perhaps a hypothesis)

A

B

Phenomenon A has no immediate causal relationship to phenomenon B

  • In a causal graph a phenomenon may have no immediate causes, one immediate cause, or more than one immediate causes
  • This is the notion of direct causality we use: A is an immediate cause of B if some change in the value of A makes the value of B change, at least under some circumstances

University of Southern California

Center for Systems and Software Engineering

5 of 15

A Scientific Inquiry

5

11/9

Ice cream sales

Shark attacks

We observe a strong correlation between the level of ice cream sales and the number of shark attacks.

University of Southern California

Center for Systems and Software Engineering

6 of 15

Scientific Theories #1

6

11/9

Ice cream sales

Shark attacks

Theory #1A: Ice cream sales causes shark attacks.

So we limit ice cream sales, expecting shark attacks to go down.

Theory #1B: Shark attacks cause ice cream sales.

So we take direct measures to reduce shark attacks, expecting ice cream sales to go down.

Fundamental problem: correlation is not causation. Need influence from 3rd phenomenon to determine causation.

University of Southern California

Center for Systems and Software Engineering

7 of 15

Scientific Theory #2

7

11/9

Ice cream sales

Shark attacks

High temper-ature

University of Southern California

Center for Systems and Software Engineering

8 of 15

Scientific Theory #3

8

11/9

Ice cream sales

Shark attacks

Shark god’s feelings

The shark god is displeased with too much consumption of sugar, and retaliates.

University of Southern California

Center for Systems and Software Engineering

9 of 15

Scientific Theory #3 (Filled out)

9

11/9

Ice cream sales

Shark attacks

Shark god’s feelings

The shark god is displeased with too much consumption of sugar, and retaliates, according to the shark shaman.

Shark shaman’s interpreta-tion

University of Southern California

Center for Systems and Software Engineering

10 of 15

Need an Ontology

10

11/9

Ontology: a description of what kinds of things (phenomena) there are in the world

  • Get different results depending on whether there are thermometers or shark shamans in the world

Causal Analysis can help by analyzing Scientific Theories #2 & 3

  • In particular, a causal analysis can identify (immediate) causes
    • Which means that a causal relationship can be identified, even though a mechanism is unknown

However, a causal analysis is limited by its ontology

University of Southern California

Center for Systems and Software Engineering

11 of 15

A Situation Where Causal Analysis Might Be Very Helpful

11

11/9

Ice cream sales

Shark attacks

Shark god’s feelings

Given sufficient data, causal analysis could tell whether the temperature phenomenon or the Shark God’s feelings phenomenon was causal.

Shark shaman’s interpreta-tion

High temper-ature

University of Southern California

Center for Systems and Software Engineering

12 of 15

Causal Analysis vs�The Scientific Method

12

11/9

A brief definition of the scientific method in this context:

  • A scientist forms a hypothesis predicting a new causal effect; the scientist performs an experiment, gathering relevant data; the scientist performs a causal analysis on the data confirming/not the hypothesis

However, it is possible for a scientist to simply gather data, and then perform a causal analysis, looking for hypotheses

  • When introduced to causal analysis, I disliked this approach, because of its flouting of the scientific method

Since then, however, I have given up my dislike, because I discovered that many scientific fields and scientists use the data-first approach as part of their normal procedure

University of Southern California

Center for Systems and Software Engineering

13 of 15

Conclusions

13

11/9

Two ways (among several) that science can advance:

  • Finding new phenomena
  • Finding new causes among phenomena

Causal Analysis can help by analyzing either of these advances

  • But a causal analysis is limited by its ontology
  • So the ontology should be acknowledged and, when appropriate, justified

University of Southern California

Center for Systems and Software Engineering

14 of 15

Idiosyncratic Annotated Bibliography

14

11/9

In preparing this presentation, I examined the Wikipedia entry on “Causality”, and found references to a few sources that struck me:

Pearl, Judea (2009). Causality: Models, Reasoning, and Inference; Cambridge University Press

    • A foundational textbook on Causal Analysis

Pearl, Judea; Glymour, Madelyn; Jewell, Nicholas P (2016). Causal Inference in Statistics – A Primer; Wiley

    • A good introduction to the how & why of Causal Analysis

"Brahma Samhita, Chapter 5: Hymn to the Absolute Truth” (2014). Bhaktivedanta Book Trust

    • Analyzes the causal nature of karma

Heller, Joseph (1961). Catch-22 (paperback). Simon & Schuster

    • Catch-22 is a rule that appears to give you a choice, but in fact only allows one result

University of Southern California

Center for Systems and Software Engineering

15 of 15

Bibliography (con’t)

15

11/9

Here is one additional citation on causality:

Smith, P. A. (2022/07/15). “The Gatekeeper” episode of RadioLab, produced at WNYC. (P. Walters, Ed.) New York City, USA.

    • A critical history of the US Supreme Court case Daubert v Merrell Dow, which decided what scientific evidence of causality is allowed into US courts

​

University of Southern California

Center for Systems and Software Engineering