analysis & reporting

The Olinger Group Analytic’s Team draws on the full range of analytic methods to exhaust the descriptive and explanatory power of every client’s data, whether obtained from primary or secondary data collection, or gleaned from the client’s existing internal sources. Years of teaching and research experience guarantee obtaining, scoring, analyzing and reporting the data that speaks most directly to the research objectives.  Whether your project requires advanced behavioral modeling, sophisticated econometric parameter estimation, or simple banners and tabs, The Olinger Group will know what you need, how to get it, and communicate it in ways you can understand, enabling you to appropriately act on results.

General Linear Model General Linear Model

A method for describing the relationship between two variables (or two sets of variables).

  • The general linear model is used to discover the relationship between independent variables and a dependent variable.  Regression analysis is used to discover the drivers of an overall concept/variable (such as satisfaction).  Not only does regression analysis help you identify which independent variables affect the dependent variable, it also uncovers the relative importance of the independent variables.

Generalized Linear Model Generalized Linear Model

This extension of GLM is employed to estimate relationships that are not adequately represented by the linear model.  This occurs when:

  • the distribution of the dependent variable is non-continuous and therefore predicted values must be similarly distributed (e.g., multi-nomial, discrete, skewed).
  • the effect of predictors is not linear (e.g., conditional, contextual).

Maximum Likelihood Estimation Maximum Likelihood Estimation

Assumes that observed values are fixed and parameters being estimated are allowed to vary until the likelihood function is maximized.  The application of these methods is an attempt to avoid bias in estimates resulting from violations of assumptions of the GLM.

Predictive Modeling Predictive Modeling

Used to discover the relative importance and drivers of overall concepts (satisfaction) and other ideas.

  • Causal modeling: Based on relational statistics like correlation-regression analysis.  Often called driver analysis, it is used to determine which items are “drivers” of an overall concept or other observables, and their relative importance in affecting the concept.
  • Path modeling: A multi-step process used when there is more than one endogenous or dependent variable in a complex process or system.  It evaluates the effects of drivers on each endogenous variable that is in turn a driver (is exogenous to) other endogenous variables in the complex process or system.  This allows for the identification of both direct and indirect effects of each variable on all other variables in the model.
  • Structural equation modeling: The most sophisticated method for modeling processes and systems.  It simultaneously does multiple indicator measurement and dimension reduction (like factor analysis) and path modeling analysis to discover the latent dimensions and their relative importance on a process of system.

Factor Analysis and Small Space Analysis Factor Analysis and Small Space Analysis

These methods test for, or identify, underlying dimensional structures in data at various levels of measurement.  R methods estimate dimensional structure in variables measuring attributes, behaviors, and attitudes.  This technique is especially useful for reducing data collection costs while maintaining measurement validity when multiple items are measuring the same thing are fail to discriminate.  It is also a theoretical alternative to ridge regression for reducing bias from multicollinearity in estimating regression models.

  • Longitudinal Analysis: The simplest form of longitudinal analysis is the tracking study which measures and reports the same thing at two or more points in time.  With only two points in time, analysis depends on conventional statistical difference tests.  This remains true when data are collected at arbitrary points in time that do not meet the assumptions of time series modeling.
  • Cohort Analysis is an aggregate method that describes cases in terms of the time of a common critical event (birth cohorts group individuals based on date of birth).  Cohorts are assumed to share attitudes and behaviors based on this commonality.  Cohort theories assume developmental models in which each cohort manifests similar characteristics as they reach the same time since entry.  Cohorts may also differ from one another in predictable ways based on their experience of differing contexts or events.
  • Time Series and Dynamic Modeling are the primary means for modeling longitudinal study measuring change over many different points of time, tends to involve a slightly smaller sample.
  • Interrupted Times Series (ITS) expands the time series model to evaluate the impact of interruptions or events, like ad campaigns or the introduction of a competitor, on the time series.  ITS models identify and describe how events impact the time series; whether the impacts are immediate or gradual, persistent or decaying, and their magnitudes.  ITS is especially useful for analyzing quasi-experimental research on naturally occurring interventions.
  • Pooled Time Series expands the basic time series models to include a cross section of units of analysis, for example locations or organization.  This increases overall sample size and degrees of freedom, allowing for the estimation of more parameters with desirable properties.  Estimates from the various pooled times series models are robust.
  • Panel Analysis temporal analysis that uses many respondents in the sample and only a few points of time measurement.  This is different from time series because it is a cross sectional method requiring a large sample size but with multiple observations in time for each case.  It can provide desirable estimates of change over time.  It allows us to discover change at the micro level with strong internal validity without the data collection requirements of long times series for each case.
  • Event History Analysis offers a number of inter-related methods for dealing with qualitative changes reported in time, such as purchase, repurchase, enroll, defect, cancel, and the like.  Real world events data generally involve left or right censoring, and this too can be accounted for during the modeling process.  This means all cases can be analyzed, not just those for which there are complete data or completion of full cycle or process.

Discrete Choice, Conjoint, and Trade-off Analysis Discrete Choice, Conjoint, and Trade-off Analysis

Complex analytical tool to gain an overall picture of the importance of different aspects.  Different combinations of “choices” are offered to the respondent over various rounds to distinguish which aspects are most important and what the ideal combination of aspects is for a product, company, brand, logo, etc.

Quadrant Analysis Quadrant Analysis

This form of analysis is used to demonstrate the position of a company in relative position to either its direct competition or the overall industry.  It allows us to gauge how the company or product is performing in comparison to its competitors and overall market.  Furthermore, we use this method to report how satisfaction upholds to respondents views of the level of importance of these aspects.