A geometrical perspective on parametric psychometric models

Francis Tuerlinckx

2024-10-16

Abstract:Psychometrics and quantitative psychology rely strongly on statistical models to measure psychological processes. As a branch of mathematics, geometry is inherently connected to measurement and focuses on properties such as distance and volume. However, despite the common root of measurement, geometry is currently not used a lot in psychological measurement. In this paper, my aim is to illustrate how ideas from non-Euclidean geometry may be relevant for psychometrics.

Applications,Differential Geometry,Other Statistics

What problem does this paper attempt to address?

This paper attempts to solve the parameterization problem of statistical models in psychometrics by introducing the concept of non - Euclidean geometry. Specifically, the author hopes to show how to use geometric ideas to study psychometric models and focus on those properties that are independent of the choice of coordinate system. This helps to provide a new perspective for understanding statistical models, especially in the fields of psychometrics and quantitative psychology.

### Main problems of the paper

1. **Insufficient application of geometry in psychometrics**:
- Although geometry is closely related to measurement, the application of geometry in psychometrics, quantitative or mathematical psychology is very limited. The author hopes to introduce the concept of non - Euclidean geometry to provide new theoretical tools for psychometrics.

2. **Invariance of parameterized models**:
- Traditional statistical models rely on specific parameterization methods, and the choice of these parameters is somewhat arbitrary. The author hopes to find a measurement method independent of parameterization choices through geometric methods, thereby improving the stability and explanatory power of the model.

3. **Geometric properties of distance, curvature and volume**:
- The author explores how to apply concepts such as distance, curvature and volume in geometry to psychometric models. In particular, the author discusses how to use the Fisher information matrix to define the distance between distributions and proposes the concept of Fisher - Rao metric.

4. **Geometric interpretation of the Rasch model**:
- Using the example of the Rasch model, the author shows how to use geometric methods to understand and interpret the parameterization problem in psychometric models. In particular, the author proposes a distance measurement method based on geodesics to solve the problem of arbitrariness in parameter scales.

### Formula summary

- **Probability of success formula** (Rasch model):
\[
\pi_j(\theta)=\frac{e^{\theta-\beta_j}}{1 + e^{\theta-\beta_j}}
\]

- **Joint probability distribution**:
\[
p(y|\theta)=\prod_{j = 1}^m\left(\pi_j(\theta)^{y_j}(1-\pi_j(\theta))^{1 - y_j}\right)
\]

- **Relative difference**:
\[
\Delta(y)=\frac{p(y|\theta + d\theta)-p(y|\theta)}{p(y|\theta)}
\]

- **Fisher information matrix**:
\[
g(\theta)=E\left[\left(\frac{\partial\log p(y|\theta)}{\partial\theta}\right)\left(\frac{\partial\log p(y|\theta)}{\partial\theta}\right)^T\right]
\]

- **Squared distance element**:
\[
ds^2=\sum_{i,j}g_{ij}(\theta)d\theta_i d\theta_j=d\theta^T g(\theta)d\theta
\]

- **Geodesic distance**:
\[
d_{\text{min}}(p(y|\theta),p(y|\theta'))=\min_{\theta(t)}\int_{t_0}^{t_1}\sqrt{\sum_{i,j}g_{ij}(\theta(t))\frac{d\theta_i(t)}{dt}\frac{d\theta_j(t)}{dt}}dt
\]

Through these formulas and concepts, the author shows how to re - examine psychometric models from a geometric perspective and proposes a more robust parameterization method.