The SurveyEngine Experiment system works within the main SurveyEngine survey system. To run an experiment, first a survey needs to be created.
Once a survey is created, experiments are authored within the system, and scenario placeholders are used to place the experiment treatments within the survey, as you might place questions or text blocks.
Data collection proceeds as per a normal survey, with experiment data being available to download as it is collected. In addition, experiment data can be modelled and analyzed from within SurveyEngine.
A number of concepts and terms are important to understand before beginning.
Attributes And Levels
The product or service for which you wish to model must be able to be described in terms of Attributes and Levels. This is necessary for SurveyEngine to understand the structure of the experiment and plan accordingly.
Attributes are also known as Product Features or Aspects, whereas Levels are the hypothetical values an Attribute could take on (and that we are interested in).
An Example:
- A car have many attributes, but for the experiment we may be interested in only 3 of those, namely: Color, Mileage and Price. Each one is independent of each other.
- There are possible values each of those attribute could take on, and that we are interested in.
- Color could have the levels of Red, Green and Blue.
- Mileage may be 100km, 150km 200km, and
- Price could be $35,000 $40,000 and $45,000.
This description of the car in terms of attributes and levels clearly defines all the possible combinations of cars attributes we are interested in. In the above case there are 27 possible combinations. From this, SurveyEngine is able to create the experiment and make predictions about the preference for all 27 cars.
Experiment Design
An experiment design is a subset of all the possible combination of attributes and levels. This subset has certain properties that allow the modelling and prediction of all combinations.
In the above example, a design of only 9 combinations of the full 27 is required to model the full set, a modest reduction in required data collection. However for larger experiments this efficiency becomes more acute. For an experiment with 20 attributes with 4 levels each the experiment design is many millions of times smaller than the total combinations.
SurveyEngine automatically produces the experimental design on request and there is no intervention required other than ensuring a design has been generated for the experiment.
Modelling
Modelling of the experiment data is the process after data collection where the observations are generalized in a mathematical form. This is done automatically with guidance from the user on what segments to model. The model itself shows the relative effect of each level on preference. However, to make predictions from the model requires a Decision Support System (DSS) below.
Decision Support System (DSS)
A decision support system, is a more intuitive form to explore the models. It allows input of any of the experiment scenarios and uses the model to predict the probability of choosing the scenario over others or even not choosing.
0 Comments