In marketing research, conjoint analysis is a very useful technique for determining preferences of consumers. Being somewhat more advanced, a conjoint analysis is trickier to get right than just a basic survey. This tutorial will guide you through the basics of performing a conjoint analysis, with a practical orientation and examples.
Why is Conjoint Analysis useful?
Conjoint analyses are performed to gain insight into the preferences of consumers. You might wonder why bother with a conjoint if you can just go up to a consumer and ask him or her directly. Unfortunately, this direct approach has multiple disadvantages. Imagine someone coming up to you and asking you about purchasing a laptop. ”Do you think the amount of hard drive memory is important? What about its screen size? Its speed? Weight?”. You probably have already understood the core of the problem: everything is important. And products are a combination of different features, not just a single attribute. Depending on your personality, you might even opt to give socially desirable answers. This is where choice-based conjoint comes in handy. You are forced to make a trade-off between the different products and attributes, and attributes are not presented as stand-alone, but in the context of the product as a whole. Next to this the implicit nature of conjoint analysis evades the danger of social desirability.
Questions this method is able of answering contain a wide range of marketing-related topics:
- Which attribute is the most important to my customers?
- What is the willingness-to-pay for the brands in the market?
- What features should the product that I’m developing have to gain a reasonable market share?
- How important is this attribute relevant to others?
To return to our laptop example: attribute refers to a key-feature of the product, like the hard drive memory or the screen size in inches. Attribute level would refer to the amount of GB’s/TB’s for the first attribute, and the amount of inches for the second.
It’s important to clarify what kind of conjoint will be studied here. (Yes, there is more than one!) Rankings based conjoint analysis focuses on ranking different products/product attributes in order of preference. It’s the most straightforward conjoint method, but it has some disadvantages. Ranking the products gives very little information on where the differences in preference stem from. Furthermore, from the results one cannot conclude the distance between the rankings, or in other words: how large is the difference in preference? Ratings based conjoint doesn’t primarily focus on rankings, but asks respondents for a product or product attribute rating. The products can still be ranked (by rating), but now it is possible to see how large the differences in preference are. These methods are not the subject of this blog post. We will be focusing on choice-based conjoint analysis. Choice-based conjoint centers on respondents having to make a choice between two hypothetical products. This is currently the most popular of the conjoint method.
Clik here to view.
Important Attributes?
Why pick choice based conjoint?
Choice based conjoint analysis has several advantages compared to the before mentioned conjoint types or a direct survey. One of its core strengths is its realism. Product attributes are not presented in an isolated manner, but combined to form a (hypothetical) product. This results in a survey which better represents real-life situations. Furthermore, it skips the problem of ”everything is important”. Respondents are forced to make trade-offs between products and their attributes. Something else that comes in handy is that you can set predefined levels for the attributes. Especially direct surveys are prone to elicit subjective answers from respondents with a very wide range. Presetting the attributes and levels makes analyzing the data a lot less complicated. Last but not least, making choices is cognitively less demanding that giving ratings. This gives choice based conjoint analysis an edge over rating based conjoint analysis, getting better quality answers.
Steps to take
To perform a conjoint analysis there are several steps to be taken.
- Define a clear research objective.
- Determine the relevant attributes and attribute levels.
- Design a questionnaire where respondents are forced to choose between products. (The products listing a combination of different attribute levels).
- Distribute the questionnaire, gather the data, and put them into a statistical software package.*
- Analyze the data.
- Make conclusions and report your findings.
*In this tutorial I will be using SAS and MS Excel.
Objectives, Attributes and Attribute Levels
The first two steps have to be taken together with the employer who hired you for your research or the responsible marketing manager. You can use a quick survey or already existing products to come up with ideas of relevant attributes and different levels for these attributes. Keep in mind that these attributes and levels will be used in the questionnaire. Therefore it should be clear what they stand for, they should be communicable and unambiguous, relevant and realistic.
Designing the Questionnaire
Designing a good questionnaire is on off the most complicated steps in a conjoint analysis. There are several requirements. You need at least two levels for every attribute. If there is non-linearity (the relationship between utility, a unit of preference, and the different levels within one attribute is non-linear), you need more than two levels. Keep the amount of levels as low as possible. Make sure that the number of levels equals that of the other attributes. Respondents will deem attributes with more levels than others to be more important! Next to this, try to set a maximum of 20 tasks for the respondent. More tasks will mean more information, but the quality of response will drop when a respondent has to answer more than twenty questions. The goal of creating a good survey is to minimize the standard-error of the part-worths. This might sound rather vague at this point, but I will explain it later on. We’re still not there yet, for there are two more demands: balance and orthogonality. Balance refers to each level of an attribute appearing an equal number of times in the questionnaire. Orthogonality refers to levels of attributes occurring independently from each other. If these demands are not met, you can end up with mixed-up effects or having too few information on certain attribute levels. Balance, orthogonality and minimizing the standard-error of part-worths cannot be done beforehand; how to take these into account will be made clear below.
For the next steps, the statistical package SAS was used. SAS comes with several macro commands that are of great support when performing a conjoint analysis. First, you need to determine the amount of hypothetical products that will be evaluated in the questionnaire, often referred to as the number of stimuli. The term ”hypothetical products” was used because these combinations of products are not necessarily a representation of actually existing products. The products are compiled in a way to maximize information gained from the questionnaire, not to imitate real products. So, if there are attribute combinations matching a real product on the market, it is unintentionally so. To determine the number of stimuli, the macro ”MktRuns” will be used. It will give you a list of number of stimuli and if they violate balance or orthogonality. How to chose one of the suggested amounts of stimuli? The lowest amount of stimuli to select is the number reported as ”saturated” in the SAS output. The maximum depend on the amount of alternatives you want per choice set. You do not want to have more than twenty choice sets, for reasons mentioned earlier. So, if let’s say you want two alternatives per choice set, 40 stimuli is the max.
Image may be NSFW.
Clik here to view.
We got balance and orthogonality covered. Still, we need to worry about minimizing the standard-error of part-worths. This is represented in the ”D-efficiency” which you will want to maximize. Evaluate the amount of stimuli you have chosen with the ”MktEx” or ”MktEval” macro. This will show you the D-efficiency in the output. Use the ”MktLab” macro to specify the number of alternatives per choice set.
Were getting close to actually putting a questionnaire on paper and sending it to our respondents. Use the ”ChoiceEff” macro to create a choice design. For this macro you also need a tricky bit of information. We need to have utility balance. Utility is something like a currency for preference. If utility is not balanced, there is little information to be had from the questionnaire. Upfront it is clear which alternative the respondent will choose, rendering the whole choice set redundant. To try and prevent this, you need to predict the valence of the different attributes. Will a higher level of a certain attribute result in more utility or preference? For example, for a laptop it is very likely that more hard drive memory will result in more utility, at least up to a certain level. It might get a bit confusing, but the perspective of the attributes is from the level you entered last. So if you have 500GB and 1000GB as levels, the perspective will be from 1000GB. The 500GB level will most likely result in less utility so this attribute gets a negative utility sign. The rule of thumb is that it is better to specify a direction as opposed to setting no direction, but it’s better to set no direction than making a wrong prediction. Use the ”Proc Print” macro to print the questionnaire design. Check for level overlap: when two alternatives in a choice set have the same level on one or more attributes. Minimize the level overlap.
Now you have a raw questionnaire. It will look like a list of numbers. Translate it into the corresponding choice sets, attributes and attribute levels. Make an appealing lay-out, write a word of thanks and such, try to use the tailored design method, and send it to your potential respondents.
Next up is putting the data into SAS and Excel, analyzing the results and reporting on your findings. Since this is already a lengthy blog post, I decided to split conjoint analysis into two posts. Conjoint Analysis Tutorial part 2 is now live!
The post Conjoint Analysis Tutorial (1) appeared first on ODMGuide.com.