A Correct Response Model in knowledge structure theory
Introduction
A toy example serves here to introduce the questions investigated in the paper. Consider a domain consisting of three items: We view each item as an elementary piece in some corpus of knowledge, or equally well as an item belonging to an examination test about the corpus. Moreover, we identify the knowledge of a person in the domain as the collection of items he masters. Because of dependencies (which may be based on the logical organization of the corpus, or just statistically observed in a given population of students), not any subset of the domain can be a state of knowledge. Suppose there are only four (potential) knowledge states, constituting the collection In principle, the correctness of the answer provided by a student to any item depends only on his knowledge state. However, variations are routinely observed in such answers. This observation motivates the introduction of stochastic effects governed by probability values. So, assume the knowledge state of a student varies in the collection according to (unknown) probabilities assigned to the states and denoted as Moreover, when a student prepares an answer to an item in his actual knowledge state, he might by inattention provide a false answer; or to the contrary, for an item outside of his state, he could by chance provide a correct answer. We thus introduce probabilities , , for careless errors in answering the three items, as well as probabilities , , for lucky guesses in those answers.
A first model (implicit in Doignon & Falmagne, 1999, page 270) defines the probability of a correct answer to any isolated item ; we call it the Correct Response Model (CRM). It first conditions the probability of a correct answer to item on the state of the student: Then, it specifies each conditional probability by taking into account the careless error probabilities and the lucky guess probabilities . In the case of the item in our toy example, the CRM sets In general, with the notations the probability of a correct answer to the item equals
A second model, the Basic Local Independence Model (BLIM), is due to Falmagne and Doignon (1988a) (see also Doignon & Falmagne, 1999 and Falmagne & Doignon, 2011). It defines the probability of each response pattern, where a particular response pattern collects the items from the domain to which a student would provide a correct answer at a given time of his apprenticeship. For instance, correct, false, correct is summarized by the pattern , (a subset of ). According to the BLIM, the latter pattern is observed with a probability equal to Both models rest on independence assumptions underlying their fundamental formulas in Eqs. (6), (7) (the formulation of the latter equation for a general pattern should be clear—anyway see later Eq. (9)). The models respectively define the probability of a correct response to each isolated item (CRM) or the probability of a whole pattern of responses (BLIM). Both rely on a fixed collection of states (some specified subsets of a domain ) together with the values of the various parameters: the probabilities of the states in and the probabilities of careless errors and of lucky guesses for the items in the domain . A particular case plays an important role in previous investigations (Falmagne and Doignon, 1988a, Falmagne and Doignon, 1988b), as well as here: by definition, the straight case obtains when for all items , that is, guessing and distraction are stochastically excluded.
Both the CRM and the BLIM are instances of probabilistic models. According to some formal process or formula, such a model predicts from any parameter point a distribution of probability values for some observables. Let us use the term predicted distribution or predicted point to abbreviate “any distribution of probability values for the observables that the model predicts”. Bamber and van Santen (2000) precisely state two fundamental questions about a probabilistic model, (i) model testability: is there any distribution of probability values for the observables that the model does not predict? (ii) model identifiability: is each predicted distribution produced from at most one parameter point? It is also common to investigate a third question, (iii) model characterizability: are the predicted distributions susceptible of an effective characterization (without reference to the underlying parameter values)?
Recently, several researchers have produced nice results about identifiability of the BLIM, the model for pattern probabilities. For instance, Stefanutti, Heller, Anselmi, and Robusto (2012) offer results on local identifiability of the model (this concept, as well as the related ones we need about models, will be precisely defined in Section 3). Spoto, Stefanutti, and Vidotto (2013) construct families of collections of states for which the BLIM is not identifiable. Heller (2017) further investigates local identifiability and so-called trade-offs among parameters. Stefanutti, Spoto, and Vidotto (2018) derive non-identifiability of the BLIM in case certain parameter transformations do exist. More recently, Stefanutti and Spoto (2020) further investigate groups of such transformations. Here, we consider the three fundamental questions for the first model, namely the correct response model (CRM). We reformulate (i) testability and (ii) identifiability, and investigate (iii) characterizability. We also point out serious difficulties for providing answers in the form of practically efficient criteria: we exhibit appropriate subcases which in other areas of mathematical research are deemed to be hopeless. Our results are often stated for the straight case. One of them indicates how to restrict in a natural way the parameter domain in order to make the modified CRM identifiable, at least for a large family of knowledge structures. Doignon, Heller, and Stefanutti (2018) is a survey chapter on the fundamental questions on probabilistic models, with the CRM and the BLIM as prominent examples. Here we give all proofs for results on the CRM, and we expose in a much simpler way how to restore identifiability.
Section snippets
Some basic concepts and two models in knowledge space theory
We briefly review the concepts from Knowledge Space Theory that we will need later on. A more detailed approach is offered in Doignon and Falmagne (1999) and Falmagne and Doignon (2011).
Definition 1 Knowledge Structure A knowledge structure consists of a finite set of items together with a collection of subsets of called knowledge states. It is assumed and .
We recall two convenient notations where is an item of the knowledge structure :
Additional axioms define particular
Fundamental questions on models
We now recall basic terminology about probabilistic models. Such a model predicts, in accordance with a general rule, the probabilistic behavior of some observable quantities from the specific values taken by parameters.
Definition 6 Model A (probabilistic) model consists of a parameter domain , an outcome space , and a prediction function with the predicted range. An element of is a parameter point, an element of a predicted point.
Definition 7 Testability The model is testable if (strict
The correct response model
According to Definition 5, the Correct Response Model (CRM) is based on a knowledge structure . Its parameters are as follows: for any state in , the probability ; moreover, for any item in , the probabilities of a careless error and of a lucky guess.
To get a formal definition of the CRM, we consider the real vector space (of all functions from to ) in which each vector has one real component , also denoted , for each state in . The simplex consists of
Testability and identifiability of the CRM
Let us first consider the CRM in the straight case, denoting again by the knowledge structure on which the model is based, and by the prediction function of the model. Thus the parameter domain is the simplex and the prediction function is To summarize, an instance of the straight CRM is specified with the notation . Clearly, the function in (15) is linear, or more adequately said: is the restriction to the simplex of a
Characterizability of the CRM
As we saw in the previous section, the prediction function of the straight CRM extends to the linear mapping (still denoted by ) The matrix of (in the canonical basis of and ) is the incidence matrix of . According to Proposition 1, the predicted range of the CRM is the -polytope having as vertices all characteristic vectors for in (that is, all columns of ). As a consequence:
Proposition 6 Any linear description of
Restoring identifiability for the straight CRM
The final paragraph of Section 4 states that, in any case, some appropriate restriction of the domain ensures identifiability of a model and preserves the predicted range: it suffices that the restricted domain be a minimal transversal of the collection of counter-images of the prediction function. In this section we find a canonical restriction in the context of the straight CRM based on a quasi ordinal space. We start with an illustrative example.
Example 5 Let and
Conclusion
The paper investigates the fundamental properties of the correct response model (CRM), a probabilistic model implicit in earlier publications on knowledge space theory (see for instance Doignon & Falmagne, 1999 page 270). In the straight case (no careless error or lucky guess), we provide in Section 5 a simple criterion for testability of the model (Proposition 2), as well as an intricate one for numerical testability (Proposition 3). We also state another intricate criterion for
Acknowledgments
The author thanks the Editor Jürgen Heller and two anonymous Reviewers for their helpful comments on preliminary versions of the text.
References (36)
- et al.
How many parameters can a model have and still be testable?
Journal of Mathematical Psychology
(1985) - et al.
How to assess a model’s testability and identifiability
Journal of Mathematical Psychology
(2000) - et al.
Note on two necessary and sufficient axioms for a well-graded knowledge space
Journal of Mathematical Psychology
(2009) - et al.
Extended formulations for order polytopes through network flows
Journal of Mathematical Psychology
(2018) - et al.
Facets of the linear ordering polytope: a unification for the fence family through weighted graphs
Journal of Mathematical Psychology
(2006) - et al.
Weighted graphs defining facets: a connection between stable set and linear ordering polytopes
Discrete Optimization
(2009) - et al.
Primary facets of order polytopes
Journal of Mathematical Psychology
(2016) A representation theorem for finite random scale systems
Journal of Mathematical Psychology
(1978)- et al.
A Markovian procedure for assessing the state of a system
Journal of Mathematical Psychology
(1988) A short proof of a theorem of Falmagne
Journal of Mathematical Psychology
(2004)