Comparing the Van Westendorp Pricing Method with Conjoint Analysis
Introduction
This is Part 2, a continuation of findings from two recent research-on-research projects we conducted involving choice of Electric Vehicles (EVs) and wireless speakers. In Part 1 we described the Van Westendorp PSM (VW) method. We also investigated order effects in VW. We found that asking the four Van Westendorp questions in high-to-low vs. low-to-high order could have a modest effect on the results. If you haven’t yet read Part 1, we recommend reviewing it before proceeding with this article: https://www.garudax.id/pulse/order-effects-van-westendorp-pricing-approach-bryan-k-orme-1vxac/
In general, at Sawtooth, we recommend choice-based conjoint (CBC) for survey-based pricing research. Yet, we recognize that there are situations in which VW might be useful. We summarize some of the advantages and disadvantages of VW and CBC below:
Two Studies We Conducted
We recently conducted two research studies with the help of PureSpectrum, a panel sample provider for the market research industry that we often work with. We’re very grateful for their sponsorship of this research!
In the first study, we interviewed 511 US-based respondents (HH income >= $30,000, age 24+) about their preferences for Electric Vehicles (EVs), which is a well-known and established category in the USA. The primary focus of the research involved a choice-based conjoint (CBC) experiment. Although we wouldn’t naturally think of using VW for pricing research studies on EVs (it’s an established market with a known price range), we decided to add the four VW questions plus the two Newton-Miller-Smith (NMS) extension questions to the survey prior to the CBC exercise for comparison.
In the Part 1 article we linked to above, we described how adding two purchase intent questions (the Newton-Miller-Smith extension–NMS) after the four VW questions allows us to derive a demand curve. We naturally may wonder whether a demand curve from VW-NMS provides similar information as a demand curve derived from CBC, both in shape and in terms of price elasticity of demand. We find in one case that they seem to line up fairly well, but in the second case they do not.
Our CBC study involved six attributes, with the following levels:
Each respondent completed 10 CBC choice tasks, showing four EVs at a time, plus the None alternative:
We estimated the standard HB-MNL model, leading to individual-level utilities.
To derive a demand curve from the CBC data, we started with a base case simulation scenario that included all 5 brands of EVs plus a None alternative where they compete for shares of preference than sum to 100%. We set the features for each of the five brands to be constant, at about average preference levels for Style, Range, and Charging Speed. Holding competitors constant, we varied the price for each brand of vehicle across the five levels of the price attribute. When a brand was not the target of the sensitivity analysis, we set its default price at $57,000, which is the average price of an EV sold in 2025 according to Kelley Blue Book. As a last step, to create an average demand curve across the brands, we averaged the five brand-specific demand curves.
We plot the conjoint and VW-NMS summary demand curves for the EV study on the same chart for comparison:
We should note that the fact that the VW-NMS and Conjoint demand curves are so similar in height (on the Y axis) is arbitrary and due to some luck. We simulated five brands at a time in the conjoint simulator plus the None alternative, averaging the results across the test alternatives’ shares of preference to create the summary demand curve shown in blue. It was arbitrary that we decided to study five brands in the CBC (we could have studied fewer or more), which would have led to higher or lower shares of preference on the Y axis.
In this article, we focus on the shape and steepness (elasticity) of the demand curves, assuming the goal is to investigate thresholds of pricing resistance, or to estimate the revenue maximization price (as well as profit-maximizing price if we had cost information). The shapes of the two demand curves are somewhat similar: both modestly convex. However, the conjoint demand curve hints at an elbow at the $75,000 price point, whereas the VW-NMS demand curve is a smoother curve.
Regarding the steepness of the curves (an indication of price sensitivity), we used the log-log regression approach to estimate the average price elasticity of demand. The summary elasticities were -2.70 for conjoint and -2.99 for VW-NMS, which is right around a 10% difference in price sensitivity. We don’t know which is more accurate, just that they are very similar estimates. (Note that these are estimates of the elasticity for a specific EV given other substitutes, not for the EV market overall.)
A clear advantage for conjoint analysis is that we can estimate brand-specific price elasticities via the market simulator by leveraging sensitivity analysis (as described earlier).
Tesla’s own price elasticity of demand estimate is relatively lower (at -1.89) than the other four brands which all group around -3. There is a core group of Tesla enthusiasts who are willing to choose Tesla even as its price increases beyond the other brands. To estimate separate price elasticities for the brands with VW we would have had to ask the VW series of questions separately for each brand.
Recommended by LinkedIn
As we highlighted in the summary table at the beginning of this article, conjoint analysis can examine many other tradeoffs that VW cannot. For example it could identify how much driving range (350 miles? 450 miles?) is enough for a 4-door sedan in the $45,000 to $55,000 price range. It can also examine the tradeoffs among charging speed and driving range. For example, if an EV offers maximum driving range (550 miles), is it going to be a significant deterrent if the charging speed is relatively slow (10-25 miles of range per hour of charging from a home plug-in connection)?
We were somewhat surprised by the similarities we observed in the demand curves as well as the summary price elasticity estimates for VW-NMS and conjoint analysis for this EV study. In the second study on wireless speakers, CBC and Van Westendorp with the NMS extension lead to different results.
Results from the Wireless Speakers Study
For a study involving wireless speakers, we included both CBC and VW (with the Newton-Miller-Smith) extension in the survey. Different groups of respondents were randomly selected to complete different sections of the survey. 197 respondents completed the CBC exercise. 398 completed the VW exercise.
With the conjoint market simulator, we generated demand curves for 5 separate brands, then averaged the results across the 5 brands to form a summary demand curve. For the VW exercise, we simulated a demand curve leveraging the two Newton-Miller-Smith purchase intent questions as discussed in Part 1 of this article.
The height of the two demand curves was not nearly so close as with the EV study, so we shifted the demand curve upward proportionally for the conjoint data for easier comparison. (Shifting the curves proportionally doesn't change their price elasticity.)
Occasionally for conjoint studies (perhaps one out of 10 or 20 projects, based on our experience), preference will start a bit lower, then increase, then fall again. When this happens, conjoint analysis is detecting a psychological price point below which there is a significant group of people who are distrusting the product, perhaps taking that low price as a signal of poor quality.
Van Westendorp embraces the idea that there can be psychological thresholds below which many respondents are distrusting the quality, even potentially leading the respondent to suggest this outcome by asking about the "too cheap" price. In Part 1 of this article, I discussed how the original Newton-Miller-Smith purchase intent extension assumed that purchase intent dropped to zero at the "too cheap" price. I also mentioned that my colleague Keith and I tend to think for most product categories this may not be a good assumption. We tend to prefer the assumption that purchase intent at the "too cheap" price should about match purchase intent as at the "acceptably cheap" price.
Because some readers might wonder what the VW+NMS demand curve looks like under the original NMS assumption (purchase intent dropping to zero at the "too cheap" price), it's displayed below in the orange line:
Changing the assumption alters the VW+NMS demand curve substantially! It introduces a pronounced kink downward between the $49 and $99 price point. Quantity demanded is essentially flat from $99 through the $149 price point, in stark contrast to the steep decrease in quantity demanded shown by conjoint between $99 and $149.
Under either assumption applied to VW+NMS, the shape of the demand curve is quite different from conjoint analysis. Which one is the truth? We unfortunately don't have access to the true demand curve to know. Conjoint analysis more closely mimics the context and act of real buying decisions among competitive offerings, so we place more faith in conjoint analysis.
Further Dissecting the VW-NMS Approach
Using the Newton-Miller-Smith (1993) purchase intent extension (NMS) to build a demand curve is fundamentally different from Van Westendorp’s original approach, which plots cumulative probability curves for the four price points and interprets the intersections. The NMS extension to VW makes it resemble other pricing research approaches (i.e., sequential monadic price tests or Gabor-Granger), but with the test price points respondent-defined rather than researcher designed. My skepticism of Van Westendorp (VW) stems primarily from this original line-crossing analysis, which doesn’t address quantity demanded at different price points. However, the NMS extension allows us to map price points to purchase intent in a way that can generate a potentially meaningful demand curve—provided certain assumptions hold.
To assess the validity of VW-NMS for building a demand curve, consider these four questions and the assumptions implicit in each:
Pragmatic Conclusion
If we accept the above assumptions as reasonable, the VW-NMS approach provides a defensible method to estimate price sensitivity and build a demand curve for a targeted product concept. For many researchers, that's a pretty big list of "IFs"! As demonstrated by our two comparative studies, we can observe differences between demand curves generated in this way versus other trusted methods such as conjoint analysis.
The VW-NMS approach doesn't mimic the shopping decision as CBC does. It's sensitive to response-style biases, purchase intent scale mapping to choice likelihoods, and the selection of price points. VW-NMS also typically is limited to estimating average price sensitivity (unless you ask the questions separately for each brand), whereas conjoint analysis facilitates brand-specific sensitivity as well as tradeoffs among other product features . Yet, for certain research situations, VW-NMS may be a practical compromise between full experimental pricing studies (like monadic pricing designs or conjoint experiments) and the incomplete original VW line-crossing interpretation that was meant to address perceptions of reasonable price ranges.
If research budget allows, we're not averse to using multiple methods to measure price sensitivity. If the results concur, we count ourselves fortunate and it gives us even more confidence. If the results differ, we place more trust in conjoint analysis.
The NMS extension dresses up Van Westendorp a bit, allowing us to estimate demand curves, but it's not as economically sound, analytically flexible, shopper realistic, and generally trusted as a well-designed choice-based conjoint analysis.
We have to remember that VW was originally devised to give an idea to a manufacturer (of whitegoods I think) of roughly what they could think about charging when they had no idea.
No one is going to argue that conjoint is a lot more robust than Van Westendorp. But simplicity, significantly lower costs and considerably faster results are pretty big advantages.
I use conjoint for most pricing work, but occasionally run a VWD to understand price perceptions (often alongside another method). I just add a big caveat: it shouldn’t define the optimal range. It’s more a read on price expectations and how much work may be needed to communicate the product’s real value.
VW is more useful at telling you what quality signals a price point sends to consumers rather than to set prices. It's pretty useless unless you anchor people to the cost of in-market alternatives. For decades The Price Is Right gameshow succeeded BECAUSE most people have no idea what something actually costs thus can't be expected to give you a useful unaided estimate of what they'd pay for it.
Very insightful. In many real-world studies, Van Westendorp works well as an initial pricing signal, but CBC tends to provide deeper insights when modeling competitive trade-offs and demand curves.