Data Science and Statistics
Insight is well known for our strong data science and statistical expertise, including microsimulation methods, multivariate imputation, modeling, weighting, and analysis of complex survey data.
Insight is well known for our strong data science and statistical expertise, including microsimulation methods, multivariate imputation, modeling, weighting, and analysis of complex survey data.
We bring extensive experience developing multilevel models to capture variation in the effects of individual versus structural or time variant components. These models help predict outcomes for situations affected by several different components—such as students, classrooms, and schools.
Our time series analysis methods assess data as occurring sequentially over time to examine trends, such as entry and exit from a particular program.
Our proven survival analysis methods help assess time from a defined starting point to the occurrence of a given event, such as the time a population receives services.
We use microsimulation to assess the potential effects of a proposed policy or program change.
We use propensity score matching methods to create comparison groups in quasi-experimental design studies. This approach helps us assess the impact of a policy or other intervention by accounting for covariates.
Our sampling statisticians are skilled in conducting complex weighting and variance estimation procedures for simple to multistage, stratified cluster sample designs using a variety of different approaches, including jackknife, BRR, and Taylor series methodologies.
We understand that using administrative program data and extant data is often the most cost-effective, appropriate, and rigorous approach to addressing research questions. Our familiarity with a wide range of secondary data sources and methods facilitates analysis and expedites turnaround times.
We often design multivariate models to predict outcomes for situations affected by more than one variable. We use methods such as correlation analysis, regression analysis, analysis of variance, discriminant analysis, factor analysis, and cluster analysis to answer varied research questions.
Insight implemented bivariate statistical testing using the statistical analysis software SUDAAN to identify differences by race/ethnicity, age, and gender. Running a series of logistic regression analyses showed the extent to which each of the variables could predict a specific behavior (e.g., likelihood of drinking and driving, probability of intervening to stop drinking and driving) or knowledge of the laws about alcohol use (e.g., awareness of the minimum drinking age).
Using MEPS data, Insight used logistic regression to identify factors that affect how individuals manage their family health information. Research results led to key design principles that inform consumer health information technology.
Insight examined measures of participation dynamics (e.g., entry rates, spell lengths, exit rates, reentry rates) using survival analysis, multivariate analysis, and time series methods. The study used data from the U.S. Census Bureau’s Survey of Income and Program Participation to examine the steady increase in zero-income households receiving SNAP over the past 15 years. Insight also conducted an event-history analysis to measure techniques used to document and examine the dynamics related to having zero gross income as SNAP participants.
Insight performed statistical analyses on the latest administration of SASS data from a sample survey of elementary and secondary schools in the United States on key topics of interest in educational policy. Analyses used replicate weights to adjust for complex survey design and time series regression to indicate broad trends.