Overview: Conjoint Research Test – aytm Help Center

Our conjoint experiment uses one of the most robust and sophisticated research methodologies (choice-based conjoint) to find the most desired combination of features for your future product, service or package out of the tens of thousands of possible permutations. You can also use it to optimize your offering for prominent clusters of your target audience.

Choice-based conjoint takes in all possible attributes and options within each attribute, creates a dazzling number of unique combinations out of them, and asks respondents to choose one package out of several presented side by side on the screen. After going through several of such screens, the model gets enough information to learn about each attribute and option in comparison to the others. Once the survey receives enough responses for the model to be satisfied, you will be able to see the importance of each attribute, the strength of each possible package, and the incremental value that each option adds or subtracts from it, allowing you to build and fine tune your perfect product or service or an entire product line.

Conjoint Express

Use Conjoint Express when you need to minimize your costs by asking respondents to respond to the minimum number of screens. Conjoint Express provides real-time analysis, and it will learn about the preferences of your entire sample group (or a subset of it if you apply filters), but it won’t be able to tell you anything at the individual level of each respondent.

Conjoint Segmentation

Conjoint Segmentation is a more expensive option, since it will ask respondents to go through approximately twice as many screens to better learn their individual preferences. When a survey is finished fielding, it will take 5-15 minutes to crunch the numbers using the gold standard of the market research industry, Hierarchical Bayesian Modeling. Conjoint Segmentation will arrive at similar results as Conjoint Express, but with a much higher confidence. It will also look into clusters of respondents for you and automagically identify different personas, if such personas emerge from the data.

Setup is identical for both types of conjoint tests, so you can choose either Conjoint Express or Conjoint Segmentation any time, right up until you’re ready to launch your survey.

In the Survey Editor

aytm's conjoint tests can be added just like you would add any of our other question types—you can simply drag and drop it from the sidebar and add it anywhere you want in the survey. You can also convert an existing question into a conjoint question type. Once selected, simply fill out the list of attributes you want to test, and the list of options you’d like to test within each attribute. Then decide how many columns to show side by side on each screen of the conjoint test; fewer columns will result in more screens displayed. You can upload photos to this question type.

The order in which items appear for each respondent can be randomized if global randomization is ON in the survey. Each specific item can be anchored to its position to make an exception from the global randomization rule.

With the Pro survey authoring package or a paid aytm membership, we give you enough space to enter up to seven attributes with seven options each, or fewer attributes with a larger number of options. As you approach the limit, we’ll show a warning on the right side of the page, alerting you to the number of remaining or extra combinations, so that you can appropriately manage your design. Please don’t hesitate to ping us if you need assistance in entering your options.

If you would like to look at the design sheet that our platform will produce for you, you are more than welcome to process and download it to analyze on your end. If you happen to have a design sheet created on another platform, you can upload it to your research test here.

We recommend running this research model with 750 to 1,500 completes. The smallest number of responses required to launch a Conjoint Express test is 400.

Respondent View

See how this question type looks for respondents

Conjoint Express and Conjoint Segmentation come with a live simulator visualization built into the stats page. Conjoint Express will populate it once there is enough information to build a model. Please pay attention to the warnings that will alert you if the sample size is too small for drawing conclusions. Conjoint Express will update the findings every time new responses are available on the page.

Since Conjoint Segmentation data analysis takes a few minutes to process, it will be initiated automatically only when the survey is fully completed and out of field. During fielding, once at least 400 responses are collected, you’ll have an option to manually initiate the analysis cycle and see interim results; we recommend waiting until the full data set is available and analyzed.

By default you will see an average package identified by the model. Click the “Best” button to roll all columns up and show the best possible combination of the considered options. If some of the options are truncated you can temporarily hide a few neighboring columns to read the full name of the option.

The importance level of every attribute identified by the model is visible in the table, and is expressed through the height of the columns in the visualization. The higher a column, the more important it is in the model and the greater impact on the desirability of the package its options will have. Sorting the table by importance will instantly update the visualization.

You can fine tune the package at any time by substituting its options manually. By clicking on an option you will either add or subtract some perceived value from the package and the model will tell you exactly how much of a trade-off you’ll be making. You can simulate the strength of thousands of combinations just by interacting with this visualization. Use your scroll-wheel or trackpad to quickly test any option in any column; the visualization will adjust itself and show you the new combination you selected.

When options have a very similar probability impact, we’ll have to hide some to keep the simulator clear. To see them all you have two options.

First, you can rollover the column and we’ll show as many as will fit in the list. When you rollover an item in the list, a black call out with a number will be rendered next to the triangle, signifying the actual location of the option on the scale, even though its label may have been pushed further up or down by other items.
Second, you can expand the table view. You can do it by clicking Expand on any of the attribute lines, or by clicking Expand All in the header of the table. You can click on an option and see the corresponding column above scroll up or down to autoselect it for you. You will see the incremental probability impacts for every option in the table, as well as the overall package strength and composition.

You can export the current view of the emulator as an image in PNG, EPS, or PDF format. You can also get the ten most desirable packages as a slide in your PowerPoint report by selecting it in the export section of the sidebar.

You can apply any combination of filters by demographics and/or traits, and have the numbers re-crunched in almost real time.

ADDITIONAL CONJOINT SEGMENTATION INFORMATION

In Conjoint Segmentation, we automatically conduct sophisticated cluster analysis of your data, and our algorithm will connect emerging subsets of the sample with other respondent information such as traits, as well as answers to other questions in the survey. This is a live customer persona generation engine, which will label personas with a hypothetical name and a photo, to make it easier to distinguish and navigate among them. Our engine approximates the most prominent personas in your sample group and shows the package that is perfect for each of them.

Please bear in mind that we’re not operating in terms of clear-cut filters. When we list the gender, age, and other traits under a persona, it doesn't mean that everyone in this cluster falls within this description. It tells us that these traits were more prevalent in this cluster, statistically speaking, and were best suited to describe the group. Another explanation or description of the persona may exist outside of the dataset, unavailable to our algorithm, so you may want to consider bringing everything you have into the survey. We’re happy to assist if you’re surveying your existing customers, for example, and would like to add your existing transactional background information into the experiment.

The icons on top let you toggle among each persona and the cumulative sample here. You can hide and expand the persona description section to manage your screen space; and of course, you can export the findings for each persona separately.

You can switch between Market share estimates and raw coefficients (also known as Conjoint Utility scores), which are used to calculate the market share, probability impact, package strength, etc. Market Share is a relative mode, helping you understand the implications of swapping any option and projected performance of the package when compared to an average package. Utility scores is an absolute mode of looking at each option and how much “power” the model assigned to it based on responses.