Isotonic regression

Isotonic regression has applications in statistical inference. For example, one might use it to fit an isotonic curve to the means of some set of experimental results when an increase in those means according to some particular ordering is expected. A benefit of isotonic regression is that it is not constrained by any functional form, such as the linearity imposed by linear regression, as long as the function is monotonic increasing.

Another application is nonmetric multidimensional scaling,[1] where a low-dimensional embedding for data points is sought such that order of distances between points in the embedding matches order of dissimilarity between points. Isotonic regression is used iteratively to fit ideal distances to preserve relative dissimilarity order.

Software for computing isotone (monotonic) regression has been developed for the R statistical package,[2] the Stata statistical package and the Python programming language.[3]

In terms of numerical analysis, isotonic regression involves finding a weighted least-squares fit ${\displaystyle x\in \mathbb {R} ^{n}}$ to a vector ${\displaystyle a\in \mathbb {R} ^{n}}$ with weights vector ${\displaystyle w\in \mathbb {R} ^{n}}$ subject to a set of non-contradictory constraints of the kind ${\displaystyle x_{i}\leq x_{j}}$. The usual choice for the constraints is ${\displaystyle x_{i}\leq x_{i+1}}$, or in other words: every point must be at least as high as the previous point.

Such constraints define a partial ordering or total ordering and can be represented as a directed graph ${\displaystyle G=(N,E)}$, where ${\displaystyle N}$ (nodes) is the set of variables (observed values) involved, and ${\displaystyle E}$ (edges) is the set of pairs ${\displaystyle (i,j)}$ for each constraint ${\displaystyle x_{i}\leq x_{j}}$. Thus, the isotonic regression problem corresponds to the following quadratic program (QP):

${\displaystyle \min \sum _{i=1}^{n}w_{i}(x_{i}-a_{i})^{2}}$ ${\displaystyle {\text{subject to }}x_{i}\leq x_{j}{\text{ for all }}(i,j)\in E.}$

In the case when ${\displaystyle G=(N,E)}$ is a total ordering, a simple iterative algorithm for solving this quadratic program is called the pool adjacent violators algorithm. Conversely, Best and Chakravarti (1990) studied the problem as an active set identification problem, and proposed a primal algorithm. These two algorithms can be seen as each other's dual, and both have a computational complexity of ${\displaystyle O(n).}$[4]

To illustrate the above, let the ${\displaystyle x_{i}\leq x_{j}}$ constraints be ${\displaystyle x_{1}\leq x_{2}\leq \ldots \leq x_{n}}$.

The isotonic estimator, ${\displaystyle g^{*}}$, minimizes the weighted least squares-like condition:

${\displaystyle \min _{g\in {\mathcal {A}}}\sum _{i=1}^{n}w_{i}(g(x_{i})-f(x_{i}))^{2}}$

where ${\displaystyle {\mathcal {A}}}$ is the set of all piecewise linear, non-decreasing, continuous functions and ${\displaystyle f}$ is a known function.

• Robertson, T.; Wright, F. T.; Dykstra, R. L. (1988). Order restricted statistical inference. New York: Wiley. ISBN 978-0-471-91787-8.
• Barlow, R. E.; Bartholomew, D. J.; Bremner, J. M.; Brunk, H. D. (1972). Statistical inference under order restrictions; the theory and application of isotonic regression. New York: Wiley. ISBN 978-0-471-04970-8.
• Shively, T.S., Sager, T.W., Walker, S.G. (2009). "A Bayesian approach to non-parametric monotone function estimation". Journal of the Royal Statistical Society, Series B. 71 (1): 159–175. CiteSeerX 10.1.1.338.3846. doi:10.1111/j.1467-9868.2008.00677.x.CS1 maint: multiple names: authors list (link)
• Wu, W. B.; Woodroofe, M.; Mentz, G. (2001). "Isotonic regression: Another look at the changepoint problem". Biometrika. 88 (3): 793–804. doi:10.1093/biomet/88.3.793.