Models

In the Models group, the aim is to improve and develop statistical methods, modeling techniques, and programs that can be used to solve thorny analytic problems in many different types of data.

For example, there are many instances in which datasets are incomplete over a long period of time. Breaks in the available data make it difficult to discern trends and understand how a set of indicators may have fluctuated over many years. Building a model that can “fill in” these missing links in the curves produced by the existing data can go a long ways toward producing more accurate assessments that can inform policymaking.

Major activities in Models:

Develop a program that can fill in missing data (Déjà vu):

This program “fills in” missing data in a time series. It is able to isolate the break in the data in a particular time series analysis and compare what happened before and after the break with instances of other data that are complete. By isolating the immediately previous and subsequent sections and comparing across many other datasets, it can approximate the missing data, thereby “completing” the curve.

Develop software to estimate disease prevalence (DisMod):

A critical impediment to the work of many researchers and policymakers interested in resource-poor settings is that they frequently have only incomplete disease registry or administrative records upon which to base estimates of disease prevalence. They may have access to limited data from individual studies, or perhaps complete data for some years but not for others. Analytic techniques can be used to account for the limitations of such data, but they are often complex and difficult to implement.

IHME is refining a set of techniques that can be used in such instances, selecting a number of indicators for which complete or partial information might be entered, and ultimately creating software that can be used easily by researchers and policymakers to produce better estimates. For instance, we have created a new version of DisMod, a model for predicting disease epidemiology originally developed at WHO. IHME will create user-friendly DisMod III software that will allow researchers to seamlessly manipulate assumptions about different indicators and see how the results vary accordingly.

Top of Page