ESRA 2017 Programme

Tuesday 18th July      Wednesday 19th July      Thursday 20th July      Friday 21th July     

     ESRA Conference App


Thursday 20th July, 14:00 - 15:30 Room: N 101

Essential Survey Data for Market Analytics

Chair Dr LinChiat Chang (Independent Consultant )
Coordinator 1Dr Sen Deng (WebMD)
Coordinator 2Mr Jochen Ertl (Kantar Health)

Session Details

Despite the plethora of behavioral data and administrative records available today, survey data continues to fill a unique and essential need in the measurement of subjective latent constructs. These latent constructs are captured in direct self-reports from customers and stakeholders about their experiences, perceptions, needs, values, attitudes, interests, preferences, lifestyle and personality attributes; and are often essential in market analytics to predict or explain choices and actions.

Market analytics on survey data draw on a wide spectrum of longstanding to innovative techniques that serve to inform and support business decisions. Solutions include market segmentation, churn/retention analysis, choice designs and experiments embedded in surveys, marketing mix modeling, pricing models, product feature testing, and much more.

For our session, we invite submissions from statisticians and data scientists who have executed advanced analytics on survey data to successfully address a real world business problem. Data could be drawn from proprietary custom surveys, or probability-based sample surveys in the public domain. We would particularly love to include projects that demonstrate the unique value of surveys in capturing critical measures not available in non-survey sources, or multi-stream analytics where survey data is integrated with ancillary sources.

Paper Details

1. A case study of order effects in responses to tasks commonly included in surveys to obtain marketing data
Dr Chung-Tung Jordan Lin (U.S. Food and Drug Administration)
Dr LinChiat Chang (Independent Consultant)

Marketing analytics are used to measure general or targeted audience’s psychosocial, cognitive, behavioral and other characteristics, which in turn can be used to understand, segment and predict audience attitudes and behaviors and to evaluate effectiveness of communication or intervention. A common tool in marketing analytics is to survey people to elicit their perceptions, knowledge, attitudes, preferences, and behavioral intentions, among other things. Often, survey respondents are asked in a survey to answer questions from a series of parts or sections, each one focusing on a specific topic and/or using a type of measurement (e.g., choices, attribute ratings). To the extent that survey data are meant to inform decisions, it is important for survey researchers to know whether and how the order of different sections in a survey may affect elicited data. In this case study, I will examine an online survey of a convenience sample of US consumers for evidence of the potential order effect. The task order of half of the respondents is: (1) to choose between two competing products of a type of food and then to provide perceptions of attributes of the chosen product, and (2) to provide perceptions of the attributes of a product of a different type of food. The other half of respondents did these tasks in the opposite order. I will apply regression analyses to the data to investigate main effects and interaction effects attributable to the task order. And I will discuss implications of any evidence of the effects for survey researchers.


2. Developing marketing personas of senior mobile phone (non-)users with survey data
Dr Ana Slavec (University of Ljubljana)
Dr Vesna Dolničar (University of Ljubljana)
Dr Andraž Petrovčič (University of Ljubljana)

This study explores the patterns of mobile phone (non-)use among seniors in Slovenia in the context of digital inclusion and active ageing. The study was carried out in collaboration with the second-largest mobile phone carrier in Slovenia, the aim of which was to develop a detailed description of different clusters of seniors according to their (non-)use of mobile phones. At the end of 2015 a representative mobile CATI survey of 1581 residents of Slovenia aged 55 years and above was conducted. The questionnaire asked respondents about their potential use of and experience with feature and smart phones, consumption of other ICTs as well as their health status and quality of life. Nine out of ten respondents were mobile phone users, among them 73% used a feature phone, while only 27% were smartphone users.
Drawing on the user-centred design approach the aim of this study was to form marketing personas of senior mobile phone (non-)users. Personas are fictional and generalized representations of archetypal users that are developed based on data collected on actual and potential users (Pruitt and Adlin 2006). Personas help marketers and designers to better imagine the users they are trying to attract and relate to them as real humans. In this study the personas were developed based on survey data using a two-stage procedure. In the first stage the respondents were grouped into five user segments: (1) mobile phone non-users, (2) feature phone users who have not heard about smartphones, (3) feature phone users who are not familiar with smartphones (but have heard about them), (4) feature phone users who are familiar with smartphones, and (5) smartphone users. In the second stage, a two-step cluster analysis was applied to each segment using a set of nominal and ordinal variables. In all five segments two clusters were computed, giving us ten clusters in total. Based on clusters a set of factoids was identified and so-called persona skeletons (i.e., lists of basic characteristics that derive from the users in each cluster) were developed for each cluster. Finally, the ten personas of senior mobile phone (non-)users were designed in more detail. We present the methodology, challenges and results, along with discussing alternative approaches that could be used to develop personas with survey data.


3. Market Segmentation of U.S. Adults Suffering from Chronic Pain
Dr LinChiat Chang (Independent Consultant)

Pain perception is subjective. It is almost impossible to compare the level of one person’s pain with another’s because pain perception varies greatly across individuals. Thus direct self-reports remain an essential means of measuring people’s experiences of pain.

Over 75 million U.S. adults experience joint pain lasting 3 months or more, defined as “pain, aching, or stiffness in or around a joint” - including pain in the shoulder, elbow, wrist, fingers, hip, knee, ankle, and toes. Some of these adults take the initiative to see a doctor for their joint pain; some do not. Some may receive an official diagnosis of arthritis, gout, lupus, or fibromyalgia; some do not. Some report being limited in daily activities due to joint pain; some do not. Some of these adults experience pain in other sites such as back and neck; some do not. Clearly, the population of adults who suffer from chronic joint pain is not homogenous. A segmentation of chronic joint pain sufferers is needed to inform patient messaging and product development in this diverse market.

Data will be drawn from a nationally representative sample of U.S. adults who responded to the 2015 National Health Interview Survey (NHIS) conducted by the U.S. National Center for Health Statistics (NCHS). The core objective of segmentation is to extract coherent clusters who share similar attributes of interest. Careful thought on input dimensions is critical because multiple segmentation solutions could be derived from the same data. Input dimensions should be directly relevant to the primary research question, and should yield maximum differentiation among respondents. To this end, statistical modeling will explore and assess coherent clusters in terms of pain perceptions, health perceptions, health insurance coverage, comorbidities, bed days, feelings of depression and anxiety, social support, sleep disruptions, exercise routines, smoking and alcohol consumption. Modeling algorithms to be attempted will include k-means/centroids clustering, agglomerative hierarchical clustering, and Bayesian model-based clustering, with cross validation to extract meaningful, coherent, and stable segments. Competing solutions will be evaluated in terms of model fit, how well they hold up to tests of replication, and how well they can support accurate and efficient targeting of patients. Segment profiles will be presented to highlight key dimensions of differentiation, with primary focus on useful and actionable attributes.

The paper will end with a demonstration of how targeted patient messaging based on the derived segments is more compelling than generic messaging to the full market.