ERCIM News No. 145 Special theme: "E-Values - Statistical Testing for the 21st Century"

ERCIM NEWS No. 145 -

Dear ERCIM News reader, ever heard of e-values?

If not, you're probably not alone. 

Michael I. Jordan opens this issue with a keynote introducing the concept:

What is an “e-value” and why has it become an object of intense study in statistics and in the allied fields of machine learning, signal processing, and econometrics? To briefly introduce the basic idea, let us consider one of the core problems in statistics – the “hypothesis testing problem” of deciding whether observed data is consistent with some particular data-generating mechanism (often referred to as a “null hypothesis”) or is better explained by another mechanism (referred to as an “alternative hypothesis”). This problem is addressed by defining some function of the data (a “statistic”) whose distribution is as different as possible under the null and the alternative. Given an observed value of such a statistic, one then makes a choice between the two distributions, doing so in a way that minimizes the probability of errors. Classical statistical theory provides a unifying framework – the “p-value” – by which the choice between the null and alternative hypotheses reduces to a thresholding procedure.

Guest editors Peter Grünwald (CWI and Leiden University), Wouter Koolen (CWI and University of Twente) and Johanna Ziegel (ETH Zurich) guide readers through this emerging framework for quantifying statistical evidence. 

The articles in this special theme demonstrate how it is already finding applications across many areas of scientific research, from clinical trials and microbiome studies to political polls, AI evaluation and financial risk backtesting.

 

<>

I must admit when I read 'e-values', I thought of an e-rendering of ethics - humanities and sciences (qualitative-quantitative : subjective-objective : ...), but as always a welcome resource.

Includes:

p.6 Introduction to the Special Theme - E-values Statistical Testing for the 21st Century
by the guest editors Peter Grünwald (CWI and Leiden University), Wouter Koolen (CWI and University of Twente) and Johanna Ziegel (ETH Zurich) 

p.8 On Testing by Imaginary Betting^
by Glenn Shafer

'Cournot’s principle does a lot of work. We predict events that have very high probability. We use the happening of events with very small probabilities to discredit theories that attribute the small probabilities to them. (Here the small probabilities are called p-values.)'. p.8.

p.37 Large Language Models as Design Partners: Automating Graphical Mockups to Refine Requirements
by Giovanna Broccia, Maurice H. ter Beek (CNR-ISTI), and Alessio Ferrari (University College Dublin and CNR-ISTI) 

p.39 Challenges in Small-Scale Medical Data Exchange Platform Development
by Hubert Schölnast, Peter Kieseberg, Patrick Kochberger and Henri Ruotsalainen (University of Applied Sciences St. Pölten) 

p.42 SCIANCE: AI for Scientific Discovery in Europe
by András Benczúr, Edina Nemeth (SZTAKI), Jonas L'Haridon (European Science Foundation) and Magdalena Brus (EGI Foundation)

My source: Peter Kunz c/o ERCIM News

^As a community mental health nurse I have a problem with the betting and gambling industries. Yes, many participants 'gamble responsibly', they know when to take a break. They understand the house rules. For a significant number of the population, they are vulnerable. Bitcoin is another example. There the 'house' is continually re-built after successive collapses. But the mathematics here in p-values, v-values and in clinical reasoning is also significant as a crucial element whether naive or formal research.

I have always thought that -

1. Hodges' model presents an implicit theory (as per many models - given their purpose).
2. It follows, that Hodges' model can also do a lot of work.

ii Learn your lines and the hyperplanes will follow

With these lines, partitions, axes and domains in mind, when a clinical practitioner is presented with a new person, whether as a patient, client, or carer ... they can, using Hodges' model (and other tools!) approach their assessment in an open and receptive manner.

This means that the information provided by the 'patient' can be readily fielded, captured whatever the context and situation.

As noted previously, my study of Hodges' model began in the late 1980s. Application in my work as a community mental health nurse, with an interest in informatics followed quite naturally(?). Primed as I was, for various reasons to carry this forward, I also carried a mathematical learning disability. At the risk of getting bogged down in my thought, use and approach to Hodges' model I need a challenge.

Mathematics is the challenge for me. It's fascinating how we have in-built 'calculators' that can help us catch a ball, and judge fairly well where to throw a ball for interception. There seems then to be an informal or naïve  mathematics, at work unconsciously. Does the same apply to Hodges' model? If so, how can I isolate, and identify it?

• Is it represented somewhere, implicit in Hodges' model itself?
• Is it (once again) to be found in the user of the model?
• Is it (more likely, and obviously) a combination of these two?
• Or, is it a product of the system, or a series of systems? 

I was reminded of what is a Sober toy, several years ago:

Is Hodges' model a selection machine?

All four original purposes of Hodges' model:

1. Person-centred, integrated and holistic care;
2. To bridge the theory - practice gap;
3. To facilitate reflection and reflective practice;
4. To support curriculum development;

- are concerned with conjunction and choice, selection. So is life itself through distinction, difference, and differentiation.

Hodges' model is a selection machine, that is both fhuman and machine driven.

A clinician may obtain the referral information through an email, a history of previous contacts can be retrieved from a clinical information system; the context and purpose supporting access to the information.   

A whole series of blog posts describe the role of Hodges' model to help assure parity of esteem across mental and physical health. What does this mean in practice?

For the practitioner, they take selected data from the referral, a history - if available, an initial telephone contact, a conversation with a colleague who remembers the person re-referred and starts to populate Hodges' model. What are the psychological concepts that arise? What are the physical?

If a referral in whatever form, or a database record can be viewed as a bag-of-words, then Hodges' model is a collection of care concepts. Four bags then. Sets or classes. An experienced user of Hodges' model may position care concepts that throws attention on the INDIVIDUAL↔GROUP axis. Lying between the INTRA- INTERPERSONAL and SCIENCES domains, this axis (like all the others) earns its keep. There is work to be done that is also of interest in machine learning:

'A support vector machine (SVM) is a supervised machine learning algorithm that classifies data by finding an optimal line or hyperplane that maximizes the distance between each class in an N-dimensional space.

SVMs were developed in the 1990s by Vladimir N. Vapnik and his colleagues, and they published this work in a paper titled "Support Vector Method for Function Approximation, Regression Estimation, and Signal Processing"¹ in 1995.' 

https://www.ibm.com/think/topics/support-vector-machine

Strange to think that perhaps the VERTICAL axis and others in Hodges' model are not precisely S-N-E-W in their bearing? There may also be several vectors at work in fact?

Image: c/o https://www.ibm.com/think/topics/support-vector-machine

The word 'naïve' has been bubbling away for a good-many years. A close colleague Silvana Bettiol, Univ. of Tasmania kindly read my draft on Hodges' model as a mathematical object, and mentioned the introduction points to Bayes theorem even if informally. Even in those initial 'clinical' encounters (and social meetings, that attend to empathy, rapport and engagement...) complex judgements are being made, beliefs tested, from what is often partial and disparate sources of information.

Checking other leads led to Frequentist and Bayesian Approaches

'Statistical inference is a series of methods used to make decisions and draw conclusions based on available data. There are two primary approaches for inference: Frequentist and Bayesian. Each framework relies on a different philosophical perspective on probability and modeling, leading to different techniques and interpretations. Each has its own strengths and drawbacks, so understanding the distinctions between them is vital for researchers, data scientists, and statisticians who aim to choose the most suitable approach for their specific analysis.'
https://www.statology.org/comparing-frequentist-and-bayesian-approaches/

More reading required and threads to run.

Earlier this week I posted re. Cromer's book -

Cromer, A. (1997) Connected Knowledge: Science, Philosophy, and Education, Oxford: Oxford University Press

Before passing the book on, p.198, Chapter 8 notes, #4:

'"Understanding" is a commonly used English word which has no precise meaning. It's sometimes taken to mean the ability to apply knowledge to new situations. In this sense, it is a very high-level skill. Benchmarks for Science Literacy says, "Learning to solve problems in a variety of subject-matter contexts, if supplemented on occasion by explicit reflection on that experience, may result in the development of a generalized problem-solving ability that can be applied in new contexts' (American Association for the Advancement of Science, 1993)." The key word here is "may." 'We really don't know how to help students develop a generalized problem-solving ability, or whether there is such an ability apart from mere knowledge of many different problem-solving strategies. Whatever the case, since we do know how to teach students to solve specific, problems. this should be the primary focus of science education' p.198.

Ack. IBM.

Learn your lines and the hyperplane will follow i

Take an empty [rectangular] space (A4 paper in landscape)

Blank! Isn't it?

Take a line. Yes, call it that.

Divide the space vertically, and equally, in two. 

Do this again but horizontally.

You can call these lines, partitions if you wish?

Or, as per adopted convention here, axes.

Now, there are four empty spaces.

These spaces can be called quadrants, planes, or domains. The latter term usually adopted here.

Labels can be decided, and assigned to the axes and the resulting domains.

Given a purpose, in practice (initially) the 'empty' spaces have the ability to assign significance to what may be placed within them.

Such decisions are non-trivial, in the sense that context and situation determine what follows, influenced by objective and subjective considerations.

The domains can contain concepts, or keywords with decisions driven by categorical reasoning.

The content of the domains can also be viewed as classes and sub-classes.

ii to follow (with revision here?)

Case formulation - Theory and Practice

 Still sorting papers, I've discovered copies of slides from the COPE course University of Manchester on Case Formulation by John McGovern.

An original purpose for Hodges' model is to help bridge the theory-practice gap.

The slides include -

• content: ingredients - preliminary formulation
• a means of understanding an individual's problems and distress in line with cognitive behavioural theory
• hypothesis to explain acquisition, and maintenance of problem
• not piecemeal

ROLE OF CASE FORMULATIONS

Bridge between theory and practice

a flexible tool for guidance

an explanatory framework to share with the patient

How types of formulation can be related to situations (Hodges' model is situated) - inference chain.

Comprehensive formulation - reflects the template that supports data gathering

inc. core beliefs - conditional beliefs

Questions for Preliminary Formulation

Main Ingredients - includes social support/relationships/employment/finances

(so this touches on determinants)

Choosing Treatment Targets

group exercise and summary

Social Sciences, measurement ... c/o Outhwaite (1991)

 

"the conceptual aspect of the subject matter of the social sciences circumscribes the possibility of measurement. . . . For meanings cannot be measured, only understood. Hypotheses about them must be expressed in language, and confirmed in dialogue. Language here stands to the conceptual aspect of social sciences as geometry stands to physics. And precision in meaning now assumes the place of accuracy in measurement as the a posteriori arbiter of theory. It should be stressed that in both cases theories may continue to be justified and validly used to explain, even though significant measurement of the phenomena of which they treat has become impossible." Chapter 3, Realism and Social Science. p.60.

 

William Outhwaite (1991) New Philosophies of Social Science: Realism, Hermeneutics and Critical Theory. London: MacMillan.

 INDIVIDUAL
|

 INTERPERSONAL    :     SCIENCES               

HUMANISTIC --------------------------------------  MECHANISTIC      

SOCIOLOGY  :   POLITICAL 

|

GROUP

quality subject/subjective

geometry, physics (time) metric / measurement / magnitude quantity *or, automated systems, robots object/objective

social sciences - humanities "Society consists of people. Groups consist of people. Institutions consist of people plus rules and roles. Rules are followed (or alternately not followed) by people and roles are filled by people.* Also there are traditions, customs, ideologies, kinship systems, languages: these are ways people act, think, and talk." p.108. Chapter 7, Conclusion. (Citing Steven Lukes.)

Institutions "There is thus an important sense in which Marx's explanandum in Capital is a conceptual question (though one whose answer required detailed empirical as well as conceptual investigation): 'why labour is expressed in value and why the measurement of its duration is measured in the magnitude of the product.' Chapter 6, Critical Hermeneutics ... p.96.

Hodges' model:
The Individual - Group(s) axis: The subjects and agents of health and social care (theory, practice, management, policy, research, history, future studies ...)

Concept analysis - Rodgers

"The method of concept analysis has not been well understood in nursing. Most significant is the observation that nurse scholars who describe or employ the method have rarely explored the philosophical foundations and implications of conducting an analysis of a concept. As a result, it is not clear precisely how the method functions or how analysis of a concept actually contributes to further intellectual progression. While the availability of a clear conceptual definition has much appeal, the actual value of a such a definition cannot be taken as self-evident." Rodgers (1989). p.330-331.

"The method entails an initial phase, a core analysis phase, and a further analysis phase (Tofthagen & Fagerström, 2010), which according to Rodgers (2000, p. 85) includes the following overlapping activities:

1. Identifying the name and concept of interest and associated expressions.
2. Identifying and selecting the appropriate setting.
3. Collecting the data to identify:
    a. The attributes of the concept.
    b. The contextual basis of the concept.
4. Analyzing the data.
5. If appropriate, identifying an exemplar of the concept.
6. Identifying hypotheses and implications for the further development of the concept."
Näsman, Nyqvist, & Nygård (2022). p.3104.

Rodgers, B.L., Concepts, analysis and the development of nursing knowledge: the evolutionary cycle. J. Adv. Nurs. 1989:14,4,330-335.

Näsman, M., Nyqvist, F. & Nygård, M. Disentangling the Concept of Well-Being in Very Old Age Using Rodgers’ Evolutionary Concept Analysis. J Happiness Stud 23, 3101–3126 (2022). https://doi.org/10.1007/s10902-022-00496-4

Book review - "Human Landscapes: Contributions to a pragmatist anthropology"

'After Ethnos', a previous book review was an enjoyable challenge. As a nurse, you are, to a certain extent several disciplines all rolled into one, sociologist, psychologist, diplomat, anthropologist ... but After Ethnos and now Dreon's 'Human Landscapes' bring their own demands.

If it was experiment that attracted me to Tobias Rees's book, Dreon's title works on two levels.

Hodges' model anticipates different 'landscapes'. The model also provides a lesson in pragmatism (another post) with pragmatism identified as a method to research Hodges' model.

On health care education programmes, how often do we write of human nature, cognition, or cognitive processes, qualitative and quantitative, subjective and objective? How quickly do we present ourselves, and the readers of assignments with dichotomies, dualities, and polarities?

Just 40 pages in they are here in expected and welcome abundance. I clearly have some reading to do [ that's why it's a review! :-) ] and from reading, much to critique in my differentiation between phenomena and as concepts presented here on W2tQ.

"... the conceptual framework for my approach to human nature is represented by cultural naturalism— that is, a nonreductive form of naturalism that assumes culture to be continuous with nature, rooted in the very organic and environmental conditions of human life, and yet irreducible to the mere association of preexisting resources." p.23

Human landscapes

"Nonetheless, even if we wish to focus on a theoretical analysis of the concept, the idea of human nature seems to display a tangle of problems, punctuated by distinctions that could be useful when treated as functional and connected to specific contexts, but which are actually turned into dogmatic oppositions. One first form of dualism concerns the opposition between the innate and the acquired, between the allegedly innate equipment existing on the genetic or neural level and the properties and ways of behaving this is assumed to give rise to. To give but one example, based on Chomsky’s influential hypothesis, the idea of a neural program for producing grammar encoded in our genes has been assumed to be an efficient (and sometimes sufficient) cause for specific linguistic practices. From this point of view, nurture, culture, and institutions seem to be something completely different from nature, since they supervene on an already fixed material substratum as an adjunct." pp.17-18.

"Taking this or that snapshot of a human could be a useful intellectual expedient in specific circumstances, but “[b]iologically all growth is modification and all organs have to be treated and understood as developments out of something else and as pointing forward to still something else” (Dewey 1985b, 32–33)." pp.19-20.

"I believe that the conceptual opposition between subjective and objective qualities is like a net with too wide a mesh, unable to catch the complexity of our interactions with the world. The problem, I would argue, is that this opposition often involves the idea that qualitative experience is a private, purely first-person experience, consequently posing problems of public access." pp.33-34.

"The section 2 will focus on the Pragmatists’ conception of experience as something broader and richer than cognition—because, in turn, cognition is understood as a specific phase within experience, responding to the needs and problems emerging from qualitative experience and ultimately returning to it. In this approach, perception through the senses and pure sense data are reframed within the context of inquiry, as features that are pointed out, discriminated, and isolated from a more holistic field of experience according to an
idea, namely, the anticipation of the possible consequences of an action or a certain assumption." pp.37-38.

Not sure why, but at one point, reading prompted consideration of: 'lived experience' and the role of peer support workers?

Starting the book, Dreon is (naturally) laying out the scope and structure of the book. Not just a preface but first chapter, and it looks like - the second. As I get more accustomed to Dreon's style (I'm a nurse remember) there are valuable lessons in laying out arguments, with more reward to follow I trust. I have a sense of a wall, not an impenetrable one, but one that is well argued - the blocks shifting into place - even if I cannot grasp their true significance to specialists in this field.

More to follow.

Dreon R. (2022) Human landscapes. Contributions to a pragmatist anthropology. Suny Press, Albany.

Many thanks to Roberta and Suny Press for my copy (confession to follow).

1+1=2

INDIVIDUAL

|

 INTERPERSONAL    :     SCIENCES               

HUMANISTIC --------------------------------------  MECHANISTIC      

SOCIOLOGY  :   POLITICAL 

|

GROUP

belief my purpose	1+1=2 0,1 ...
society social groups social networks	1+1= ? "When people talk about how math is or should be apolitical, because "1+1=2" does not depend on politics and neither does the Riemann hypothesis, it seems that the only math that they can think of is pure math. This line or argument completely falls apart for applied math."  @imathematicians [thread]

My source:

https://twitter.com/imathematicians/status/1530724466073952261?s=20&t=eF1XRD9HbkSJ2XWWbIk2qQ 

Previously: 'maths' ,  '2 + 2 = ?'

https://twitter.com/imathematicians/status/1530724466073952261?s=20&t=eF1XRD9HbkSJ2XWWbIk2qQ

Book Review: [ii] An introduction to Health Research Methods

Where were we from Part [i]? For the book contents listing look no further. The publisher's site also provides sample pages.

For part ii we are up to chapter 14. Correlation studies, also known as aggregate or ecological studies had me scurrying to the index. While such studies usually rely on secondary analyses I thought about the power of social media to disrupt such distinctions. 'Social media' is there p.165 and this would be primary 'live' (the idea of a feed sounds 'live', but then web crawling can be quite retrospective) or secondary and might deserve a chapter of its own in. Despite such queries figures proceed to inform the analysis of correlation pp.97-100.

At step 3 we are still under a third way through  the text, excluding the glossary and index. Chapter 15 explains research timelines and protocols. While trying to formulate my own research question, I created a research plan and Gantt chart (figs 15-1 & 15-2). This prompts me to revisit the exercise as an independent scholar with the decade long aspiration to create a new website. The treatment here is aimed, of course, at research teams with responsibilities and a principal investigator. Figure 16-1 on sample populations and how they are derived I sometimes share with students on placement as I encourage them to be researchers. They should always be asking questions, as students, at their job interview and a new post provides an ideal opportunity to hear problems, issues, uncertainties. Many managers miss this - a fresh pair of eyes.

Vulnerable populations deserve and are given specific mention. Chapters 17 and 18 covers primary research concepts, such as, sample size, power estimation, reliability and validity. The systematic approach throughout the book is evident here. My initial note about lack of whitespace sees Jacobsen advocating for whitespace in questionnaire design. This advice flows into surveys, interviews and other assessments. The research essential of Boolean search is described with the process of a systematic search strategy, also allied with Forest and Funnel plots. Do take time with chapter 25 on writing research grants, even if you are an individual. It is obvious, but your financial situation can greatly affect your staying power in a programme of study that incorporates research, or is research led. Personal circumstances can change.

In step 4 if research has its quarry face, then we meet it here in (what can be) the hard-grind of data management (chapter 26) code books, data entry, data cleaning and data coding. The maths is invariably explicit in chapters 28 and 29 on descriptive statistics, regression analysis. Here the figures and text really do go hand-in-hand. If baby should never be in the corner Step 5 reminds us that research should never just be stuck on the shelf. In reporting findings you will find posters, conferences, presentations and writing papers, citing sources, and seeing an article through submission, revision to publication. A chapter deals with success - or how to keep going which brings readers to end with a glossary that is not cross referenced but the index is thorough.

This is very good introduction and the title really reflects the aim and content. This is an excellent primer that you can return to after the paint is well dry. It will be a pleasure to pass this book on to a nurse working in research - clinical trials. As is often the case there is an online access code to additional course materials. I have not broken the seal but may add further comments if they follow. There are always other things that might be included. As research stresses literature is there a way to include literacy, to wrest health literacy from the fog of the research question? Visual methods are increasingly used and the comic format used for educational, and even presentation of theses and dissertations. Genomics and the advent of (medically) personalised care may demand additional knowledge for researchers of all levels.

The other thought I alluded to in part [i] does not affect my recommendation for this book; but it concerns the primary, secondary and tertiary forms of research, literacy and access to information. Yes, these have an established place in the research process and lexicon (inclusion/exclusion criteria, informed consent...) with a history to reinforce this foundation. But is there a problem as this research is supporting health systems that really must change. Does something further in research need to change? Where and how could we incorporate the quaternary view? Never mind a chapter: could that have a step all its own?

Part [i]

With thanks to Emma Phillips, Marketing Manager: Academic Class Learning.

Book Review: An introduction to Health Research Methods [i]

For this review, the preface says it all in the purpose of this book as a: "necessary foundation for meaningful improvements in clinical and public health practice." From an outcomes based view, to the research process itself the book is organised through five - very recognisable - steps:

1. The Research Question 5-43
2. Study approaches 35-100
3. Study design and data collection 101-211
4. Analysis 213-271
5. Dissemination 273-345

I have included the page numbers across the steps to give you an idea of the weighting. The text cuts across the various methods, disciplines and credentials of the researcher.

If you are looking for whitespace this book is not for you. As a result though this adds of course to value-for-money. As evidence of packing-it-in the publishers have packed the pages, which for a research book is welcome. The execution while dense with ink on the page, is well delivered in presentation. The font is large, clear, distinct and readable. The adopted format is established very quickly. The first paragraph refers to a figure which is also on the first page. Figures are used to good effect throughout the book, shaded tones not colour.

As per the steps outlined above, two structures to help formulate a general research topic and question are described Exposure, Disease, Population [EPD] and PICOT. Mental health represented here encouraged my reading early on. Literature is also explored, the process of reviewing and systematic revires and meta-analysis in chapters 3 & 22.

With so much research being conducted the matter of originality and gaps in the literature are included. Figure 3-1 provides useful "Ideas for new studies" p.18, the issue of gaps also earns a figure of its own. Chapter 4 quite rightly brought pangs of guilt here dealing with "Focusing the research question" - primary (new data), secondary (existing data), tertiary (review and synthesize the literature). I have struggled to finalise a goal and set specific objectives. This did give me something to ponder upon which I will save for the end.

Each chapter begins with a pithy sentence of advice. Some chapters are very short, e.g. chapters 5 and 6 but this is justified and not suggestive of a need for an alternative structure. Step 2 signifies the arrival at more technical content; time frames, population selection, case definitions, cohort studies, cross sectional surveys, case control studies - with odds rations. This is all framed in accord with the introductory theme. Even an introductory book cannot avoid maths and Jacobsen deploys the figures to very good effect, for example Incidence in figure 11-5. Chapter 12 may be less heavy on the maths but it is a vocabulary expander for the uninitiated in research, as figure illuminate random control trial approaches, randomization efficacy, numbers needed..., flows of participants and screening a diagnostic test results (positives and negatives). The author's picking out Qualitative Research - Chapter 13 gave me hope for Hodges' model which can readily differentiate between quant and qual. And yes, brainstorming and mapping are mentioned at the start of the book.

TO BE CONTINUED...

Fallding: 'model'

"The ideal type, as a case of the heuristic, resembles what is now called a model. ... He [Nagel] makes model refer simply to the mental imagery with which we clothe our theoretical entities in order to have some pictorial grasp of them : ... essentially analogies, yet they do more than help us visualize; they help our thinking to unfold by taking it in new directions. 

[Model] is taken to mean a system of concepts that is useful in mapping the variables in a field under investigation. It may be a corner of a field or the whole field of the subject: a model may be of bureaucracy, for instance, or of society in general. ... in scientific work we do not start out with an empty mind, nor even with isolated hypotheses. We start with a guess about the whole structure of our universe of discourse. The coherence that goes into this system of concepts, a coherence whereby each concept is defined by the relationship in which it stands to the others, is analogous to explanation but is not explanation in fact. For it is not an account of reality as it has been experienced" (p. 510).

Harold Fallding. Explanatory Theory, Analytical Theory and the Ideal Type. In. Thompson, K. and Tunstall, J. (eds.): Sociological Perspectives, Penguin Books, Harmondsworth, 1971.

Excerpt from Harold Fallding. (1968) The sociological task. Prentice-Hall.

Is Hodges' model a Self-Organising Hypothesis Network?

"The proposed approach is based on 3 key steps. First, derive knowledge from different sources of information (experimental data, expert learning, machine learning). Secondly, unify the different knowledge representations. Finally, organise the unified knowledge in a way that captures its generalisation hierarchy and facilitates the design of efficient prediction algorithms (Figure 3)." page 2 of 21.

Figure 3 The SOHN methodology. Different sources of knowledge are unified using a common representation based on the concept of hypothesis. The hypotheses can be organised into a hierarchical network to capture the knowledge in a standardised way (page 3 of 21).

Straight away this figure in Hanser et al. (2014) suggested learning and learners and the four care domains of Hodges' model. There are profound differences of course.

The information at the start takes the form of concepts, data, but 'information' also applies. The context in combining chemistry and informatics is radically different when it comes to who, or what is learning in that first arrow. The paper's keywords are:

Machine learning Knowledge discovery Data mining SAR QSAR SOHN Interpretable model Confidence metric and Hypothesis Network.

To a healthcare or social care student the knowledge encountered may well be raw. The principle of Hodges' model encourages the generation of hypotheses within and then across the model's knowledge domains. The learn-er can then attempt to unify these thoughts and reflections, organising them in a variety of ways. Hodges' model is not itself in self-organising, but there is a pattern-learning process for the learner and a pattern-matching process for the expert or specialist.

Hanser, T., Barber, C., Rosser, E., Vessey, J., Webb, D., & Werner, S. (2014). Self organising hypothesis networks: A new approach for representing and structuring SAR knowledge. Journal of Cheminformatics, 6(1), 1-21.