ePrivacy and GPDR Cookie Consent by Cookie Consent
   
AAPOR
The leading association
of public opinion and
survey research professionals
American Association for Public Opinion Research

Total Survey Error Meets Big Data

Click here to purchase this webinar kit.

Member Price: $175.00
Nonmember Price: $235.00

Student Pricing Available

Extending the Total Survey Error Perspective to Multiple-Surveys and Big Data
About This Course:

The total survey error (TSE) paradigm was originally developed to apply to single surveys. As such it is both a theoretically and an empirically powerful approach to identifying and reducing survey error and thereby increasing reliability and validity. IN this presentation, TSE is applied to multiple surveys and to the use of auxiliary information from so-called Big Data and other sources.

Learning Objectives:
  • Learning about the total survey error (TSE) paradigm.
  • Applying TSE to multiple surveys.
  • Incorporating Big Data into the TSE paradigm.


A Total Error Framework for Generic Datasets and Estimates
About This Course:
The course reviews the familiar total survey error framework and discusses how this framework can be generalized and extended for generic probability and nonprobability datasets. The framework considers all the major sources of error and their associated mean squared error components and how they combine to produce an expression for the overall accuracy of a data set mean. Because of its generality, this framework is particularly useful for comparing estimates from surveys with estimates from massive data sets, often with surprising results. An example based upon real data illustrates how the accuracy of estimates from an administrative data set covering 80% of US households may be inferior to a sample survey of 6000 households with a 55% response rate.  The example will also illustrate why error mitigation to improve the comparison may be futile in many situations. Finally, the implications for evaluating the accuracy of hybrid estimates based upon integrated survey and non-survey (Big Data) data sets are discussed.

Learning Objectives:
  • The processes by which data are generated and errors are introduced in generic data sets and estimates.
  • The concepts of data encoding error, sample recruitment error, their associated subcomponents and their relative importance.
  • Understanding what information is needed to evaluate total estimator accuracy and to develop mitigation strategies for improving accuracy of estimates from nonprobability samples.


The Mixology of Samples - Blending Probability and Nonprobability Samples
About This Course:
The times, they are a-changing! Subsequent to the webinar conducted in February on the fundamentals of nonprobability samples, this presentation provides concrete recommendations for producing inferences from surveys that rely on blended probability and nonprobability samples based on our years of hands-on experience. After a brief recap of the previous webinar, we will discuss robust weighting methodologies that can reduce some of the inherent biases of nonprobability samples, including extending weighting adjustments beyond standard geodemographic realignments. We then introduce an option for the optimal integration of surveys from different samples, and touch upon fusion of ancillary data from external sources. Next, we will provide recommendations for practical alternatives for approximating error margins from blended samples. We will end this webinar with some contemplations about the new directions in survey sampling in the digital age.

Learning Objectives:
  • Assessment of current practices for weighting nonprobability samples
  • Recommendations for optimal calibration and blending of survey data
  • Practical options for approximating error margins


Data Science Trends and Tools for Measuring Attitudes and Behaviors
About This Course:
In its broadest sense, “data science” is an interdisciplinary field about scientific methods, processes, and systems to extract information and insights from data in various forms – designed or organic, structured or unstructured. Working across a wide range of topic domains, data science approaches are now used routinely in the commercial sector for measuring or drawing conclusions about people’s attitudes and behaviors and becoming more commonplace in other sectors as well. Operating parallel to or in conjunction with more traditional statistical survey research approaches, data science is having a transformative effect on the ways in which researchers seek to understand attitudes and behaviors. In this webinar we explore some of the major trends in data science as it pertains to the measurement of attitudes and behaviors as well as its relationship with traditional survey approaches. We then discuss at a conceptual level some of the key tools and approaches being used and how they are being applied in the field. In doing so we take a closer look at three of the more widely used areas: machine learning, text analytics, and data visualization. We conclude with a look at where the field of data science is going and its longer term implications for how we measure and understand attitudes and behaviors.

Learning Objectives:
  • Understanding the field of data science and how it relates to traditional statistical survey research efforts.
  • Identifying some of the key methods and tools being used in the field and how they might be applied.
  • Delineating some of the key advantages and disadvantages of data science approaches for understanding attitudes and behaviors.