How to use Metrisim's experimental design module for conjoint analysis

This module generates an experimental design that you can use to create product profiles to show respondents in a survey.

The type of design typically used in conjoint analysis is a fractional factorial. It is called a fractional design because on a fraction of the possible combinations are shown to respondents to reduce work load. You don’t have to worry though since a model is used in the simulator that allows us to estimate the impact of combinations not shown to the respondent. Typically an algorithm is used to find such a design by optimizing a criterion such as d-efficiency; as is the case in Metrisim.

This is the only file required as input for the design module to work. This tells the design module what attributes and levels define products in your market.

You won’t be able to cover all attributes usually, unless it is a simple product, and so you should focus on the attributes and levels that are both important in your view, and simultaneously are within your organization’s power to alter.

For instance, in the case of the UHT milk survey - used as a demo within Metrisim - it was decided that all fat grades (full cream, skim etc.) would be offered on the market; and so there was no point in including fat grade as an attribute. Instead respondents were told to simply assume they were buying their preferred fat grade. The package images were even altered to remove any references to fat grade to avoid having to vary this across product profiles.

As described in more detail in the setup files tutorial, place the attribute names in the first row, and under each, its corresponding levels. Save as a csv file, preferably UTF8 if the option exists in the software you are using.

Once you login you will automatically be directed to the simulator, as you will tend to spend more time here than with the other modules. So select the menu and then select Designer.

If you haven’t already, upload the attribute_map.csv.

Metrisim’s designer module automatically estimates the minimum number of product profiles (runs) needed. However it is usually advisable to add a few extra runs; you can indicate how many using the slider. This is especially necessary if you wish to make use of the respondent R squares for quality control in the estimator module later on.

Click run to start. Metrisim uses a Federov algorithm to find the best possible D-efficient design within a reasonable amount of time. It does this using the dummy coded design matrix for a full factorial as a starting point. It is important that this is done on the dummy coded matrix and not a categorically coded matrix (as the final design_text.csv is) as there is typically no quantitative meaning, even in the case of price which has its own special treatment in a metric conjoint analysis context - where instead of a single partworth being estimated for price, a partworth is estimated for each price point, which is then used in conjunction with interpolation in the simulator.

You can select download and then click on the link that appears to download a zipped file containing the design and another file indicating the correlations between the level dummy codes. Ideally you would prefer designs with lower correlations between the levels of different attributes; while large correlations between levels of the same attribute are to be expected as they are mutually exclusive.