Editor’s note: Michael Lieberman is founder and president of Multivariate Solutions, a statistical and market research consulting firm.

Conjoint analysis is in the “trade-off” family of market research techniques. It has proven to be profoundly useful in helping marketers shape new products, determine maximum levels of product enhancement and predict market share. In addition, conjoint analysis can be used to identify the best advertising message by revealing the features that are most important in product choice.

Essentially, trade-off analysis allows the researcher to throw all of his or her options - such as various features, price ranges, brand names, etc. - into a carefully constructed questionnaire. Respondents are asked a series of product purchase interest questions. The data are then run through the black box of conjoint procedures that yield mathematical results called utility scores. These models allow the researcher to simulate the marketplace in great detail with surprising accuracy.

In a real purchase situation consumers do not make choices based on a single attribute. Consumers examine a range of features or attributes and then make judgments or trade-offs to determine their final purchase choice.

This is as true with the “choice” made in political situations, such as assessing the viability of a candidate, determining the underlying roots of issue support - such as gun control, housing reform or budgetary management - or choosing a tactical strategy when pushing legislation. This analysis, when used properly, can provide insight to the governor when he sets policy priorities or assesses main concerns dominating voters’ minds as the election nears.

This article will review two constructs of conjoint that are specifically suited to political applications. These particular conjoint approaches can be administered over the phone or on the Internet, which keeps costs and timing manageable (very important in a flash poll). Also, the results can be easily filtered by key voter groups in order to compare them to each other and assess the relative importance of issues within each group.

First model: assessing levels of preference

The first model looks at different levels of the day’s key issues. Its goal is to measure the preference between each issue, and then construct a simulation model allowing political planners to find the optimum policy pie or the cost of circumstances on approval ratings. In this case we are going to be looking at three key voting groups across three political issues in the imaginary state named Utopia. They are summarized in the first table below.

For each issue, varying levels are tested.

Unemployment

  • 2 percent
  • 4 percent
  • 6 percent

Property taxes

  • Low
  • Medium
  • High

Utopia state budget

  • Balanced budget
  • State budget deficit of $5 billion
  • State budget deficit of $10 billion

In order for the conjoint analysis to work, a computer-generated plan for the survey is run to make the outcome statistically viable. In the above case, with three issues each containing three levels, the respondent would be asked to rate nine choice scenarios. The second table above is an example of how a few choice scenarios might look:

The survey can be done in two ways. The first is to present the respondent with each of the nine scenarios and ask him/her to rank them. However, this can be difficult over the phone - it is a challenge for a person to hold nine concepts in his head - and requires special Web programming in order for all nine scenarios to be seen on one screen.

The second method for administering the survey is to ask the respondent to rate the state government given the following conditions on a finite scale, say 1 to 10. A sample question would look like this: “On a 1-to-10 scale, how would you rate the Utopia state government under Governor Bob Perfect if unemployment is at 2 percent, property taxes are high, and the budget deficit is at $5 billion?”

In our example, each respondent would be asked to rate nine similar questions.

The output has two levels, the first is an importance for each issue. Next, a second utility score for each level of each attribute. The first gives us the relative importance of each issue, the second allows us to gauge how much support would drop if inflation went from 2 percent to, say, 6 percent.

Figure 1 summarizes the aggregate results of the importance of each issue broken down by voter group.

Evidently, younger people are more concerned about finding work and have a relatively higher concern for balancing the state budget than older voters. For working adults, unemployment and property taxes are equally troubling, while seniors want to pay low property taxes.

Figure 2 shows the utility scores for each attribute for the entire population - the conjoint procedure also generates utility scores for each voter group though they are not shown here. There are two ways to utilize these numbers. The first is to eyeball them and assess the relative strength of each of the utility scores. Obviously, for the whole population, the incremental loss of utility due to higher property taxes is the primary concern.

Another useful measure is to see what the relative loss of support would be, given a certain event. For example, the top utility sum - 2 percent unemployment, balanced budget and low property taxes - is 9.38. If local property taxes rise from low to medium due to state budget cutbacks, the utilities sum drops to 7.23 - a loss of about 23 percent. This translates directly into support lost for the governor.

Obviously Governor Perfect does not want that to happen. If unemployment stays at 2 percent and property taxes stay low, but the governor has to operate with a $5 billion deficit in order to help municipalities keep their taxes low, the score goes to 9.06 - a loss of 5 percent. It’s clear the governor should undertake deficit spending in order to keep property taxes down. He should also be sure to keep unemployment as low as possible. That affects him as well. A jump of 2 percent in unemployment means a drop in utility of around 18 percent.

Second model: shaping the message

In the second variation on conjoint, all issues that are included have only two levels: either included in the campaign/speech/advertisement, or not. For example, a candidate has six campaign issues that he has specified. Which of those should he emphasize in his campaign? Or, during his State of the State speech, in what order should Governor Perfect present the key challenges facing Utopia?

Below is a list of six issues that might appear in a gubernatorial/senate race or State of the State address:

  • ending runaway development in Utopia;
  • reducing traffic congestion on Utopia’s highways;
  • improving the quality of teaching in our classrooms;
  • gun control;
  • protecting the environment;
  • improving Utopia’s economy and increasing the number of jobs in the state.

For six issues the computer generates eight choice scenarios. An example question might be, “If a candidate’s main platform were gun control and protecting the environment, how likely are you to vote for him?” Or, “In his State of the State address, Governor Perfect plans to emphasize 1) ending runaway development in Utopia, 2) improving the quality of teaching in our classrooms, and 3) protecting our environment. How important would it be for you to hear his speech?”

The output would look like Figure 3, which is a summary of relative importance scores once the analysis is complete.

Here we are concerned only with the importance of each main attribute. The conjoint analysis reveals that Utopians are concerned more about traffic, the state’s economy and the quality of teachers. The environment and gun control, in an urban state like Utopia, are low down on the list.

This variable can be also be filtered easily by key voter groups, so when the governor goes to speak, say, in front of a group of Democratic women, he can quickly filter the analysis, rerun the program and get utility scores for this group’s preference of issues.

Applications for advertising research

The applications of conjoint analysis described above can be easily used in advertising and marketing research simply by changing the labels from “candidate” to “product.”

In the first example, instead of using the conjoint analysis to set action plans for Governor Bob Perfect’s Utopian State agenda, it can be used to plan product manager Bob Perfect’s newest toothpaste. “State Unemployment” becomes “Tube Size,” “Property Taxes” becomes “Flavor,” and “State Budget Gap” becomes “Box Design.”

In the second application, instead of shaping the message for the governor’s reelection, you could design a magazine cover by testing the presence or absence of logos, or the impact of colors or article teaser font sizes. Trumpet the preferred advantages of your bank’s credit card based on what is important to the consumer. Or advertise your client’s newest cellular telephone based on sound research.

Crystal ball

Not every dollar spent in political campaign (or consumer product) advertising is created equal - some of those dollars generate far more impact than others. Indeed, it seems that if one only knew beforehand which dollars to spend, and at whom they should be directed, success would be far more certain. Conjoint analysis can be, in large measure, that crystal ball.