Editor’s note: Robert Roy is vice president and general manager of the Chicago office of Total Research Corp.

Have you ever tried to fit a square conjoint program into a round problem? (What does he mean by that?) All right, I'll tell you what I mean. Conjoint programs have certain expectations. Among those expectations are these:

-Every level of every factor can be paired with every other level of the other factors. (A mouthful, eh?)

-If it's full profile conjoint...

-the profiles will be rated, sorted, and rank-ordered.

-the profiles are realistic to the respondent. Strange products do not appear.

Now let's hear a big, "Yeah, and so what!" The "so what" is that in real life it can often be a tad difficult to meet these limitations. Recently, however, there has emerged from the academic woodwork discrete choice modeling.

The good news

Discrete choice modeling does not force the pairing of all attributes. Therefore, unrealistic products are not produced. The respondent does not rate, sort, nor rank-order, but acts as if in the marketplace. The respondent selects or chooses which product to buy. This freedom from the tyranny of standard conjoint was brought to us by Jordan Louviere at the University of Alberta. For those interested in the mind-splitting details, I recommend the Sage Book publication number 62. Everyone else can just read on.

An example

Discrete choice theory started in transportation modeling. Usually, prices of the modes of transportation are considered, along with travel times and waiting times. But my purpose is exposition, not the design of a definitive transportation study. So, with meager apologies to the transportation experts, let's say we're studying the behavior of those who travel to work in the Chicago downtown Loop area.

While we're assuming, let's say we have a probability sample of such folks that numbers 1,000 commuters. By definition, our sample will represent all such commuters (+ or -). So much for the sample. Isn't this fun? On to the interview.

Experimental design

In Chicago, commuters can travel to the Loop in at least four ways. Their primary mode of transportation could be by:

1. Private car
2. Bus
3. Train
4. EL (The CTA elevated train)

They can also select some other mode such as walking, riding a bike, finding a comfortable camel, etc. In our simplified example we will only consider the cost of each mode of travel.

Below are listed costs that could be associated with each mode.

I. Gasoline price
      1. $.90 per gallon
      2. $1.45 per gallon
      3. $2.00 per gallon

II. Cost of all day parking
      1. $7.00
      2. $9.80
      3. $13.75

III. Bus rate

   
Zone A
   
Zone B
   
Zone C
      1.
$2.00
   
$2.50
   
$3.00
      2.
$3.00
   
$3.75
   
$5.85
      3.
$4.00
   
$5.00
   
$7.85

IV. Train rates

   
Zone A
   
Zone B
   
Zone C
      1.
$3.5
   
$4.75
   
$8.70
      2.
$4.75
   
$6.40
   
$11.70
      3.
$6.00
   
$8.10
   
$14.80

V. CTA El rates
      1. $.75
      2. $1.15
      3. $1.50

In this example, there are five (5) cost factors, and each could assume three (3) separate levels (35). An experimental design with 16, 18 or 27 combinations of these price levels will allow us to measure the impact of each price on the transportation mode selection process as well as the interaction of the price of gasoline with the cost of parking.

The task

Here's the respondent's task. And simple it is, too! A transportation scenario, or situation, is presented. The pieces associated with each mode of transportation are given. The respondent states which he would select, given the prices. He (or she, I know, I know) may also say that he would elect to travel by some other mode (such as the camel).


Figure 1

Is this easy or what? The respondent doesn't have to rate, sort, or rank-order. Instead, the traveler does what is done in real life-selects an option. Each respondent evaluates several scenarios (perhaps as many as 27). Or, we may want any one individual to evaluate only nine or so situations.

Descriptive data

What do we get from our 1,000 interviews? First, and obviously, we get the frequency with which each transportation mode is selected under each scenario. That, in turn, allows us to estimate the relationship between the frequency of selection, the cost of travel for that particular mode, and the costs associated with the other modes (Louviere outlines how to do this). So the analysis is at the group, not individual, level (Louviere points out that in some cases, individual level analysis is possible).

As with conjoint, parameters (utility scores) are derived. These parameters can then be charted to better understand preferences. Having said that, let's go back to the example, invent a few parameters, and chart away. (See figures 2 and 3.)


Figure 2



Figure 3

Market model

Also available is the possibility of constructing an interactive computer model that operates on a PC. With the model, the "what if" begins. For example, Ed Goodride, the acting director of Second City Bus Service, might have burning ambitions to be promoted to full director. He has considered the usual tactics of monetary grease, slander, and obsequious behavior. Quite naturally, he's tried all of that and one or two other things. But now he's desperate and is willing to resort to data analysis! With trembling lips he forms the questions.

"Well, gosh all get out and gee, I wonder what might happen to ridership and revenue if..."

  • rates were increased to the $4.00/$5.00/$7.85 plan?
  • while Second City increased rates, gasoline increased to $2.00 and the cost of parking averaged $10.50?

Ed's questions do not go unanswered. His friendly market research analyst has conducted the survey, analyzed the data, and now has the PC model. Ed can quickly and easily see the sensitivity of commuters to a price change for Second City Bus Service, and to price changes for competing modes of travel. In a couple of hours he has arrived at a strategy. With his boss out of the office and no one else around to impress, Ed engages the analyst in conversation.

Ed: What did you say this technique is called?

Analyst: Discrete choice modeling.

Ed: That's a mouthful. I hope it's not indiscreet. Ha ha. (Big joker, this Ed.) So why did we do it this way?

Analyst: Do you want me to chit chat, or just list the reasons? Ed: Do I make it a big habit to talk to you? Just lay it on me! Analyst: OK. Ya see, this is superior to standard conjoint.

Ed: (interrupting) To what!?

Analyst: Never mind. Here are the advantages:

1. The task is more realistic for the respondent.

2. Analysts don't have to translate ratings into behavior.

3. Current product offerings are easily studied.

4. Problems too difficult for conjoint, without heroic assumptions, can be handled.

5. And get this! We do not have to assume that the market never changes in size. This approach shows us the expansion or contraction likely to occur.

Resolution

With this strategy in hand, suggested by discrete choice analysis, Ed proceeded to campaign for the directorship of Second City Bus Service. On a wall somewhere in the Loop the following was written:

Every good researcher revels when he's joining his factors and levels but the customers howl when the combos are foul and they set out to Iynch the poor devils *

*Thanks to Bob Bisciglia*