Geek looking at data

A Tribute to David Huff: The Man and the Model

Sep 4, 2014, 05:13 AM by Environics Analytics
The field of applied geography lost one of its giants a few weeks ago. And while others will talk about the impact of David Huff the man.

The field of applied geography lost one of its giants a few weeks ago. And while others will talk about the impact of David Huff the man, I’d like to pay tribute to him through one of his creations that revolutionized business, planning and decision theory: the eponymous Huff model.

First some background: Early in his career, David was a business and marketing professor at UCLA. But he was not a typical marketing professor because of the unusual eclectic training he received at the University of Washington during a critical time, the dawn of the quantitative revolution in the social sciences. His training was strongly influenced by Bill Garrison, Brian Berry, John D. Nystuen, and Duane Marble – all outstanding early analytical economic geographers. He was also influenced by early quantitative sociologists, game theorists and rational choice theorists like R. D. Luce and Howard Raiffa. He emerged from this exciting time in graduate school as a holistic quantitative social scientist. His thesis work from 1959-1962 in modeling trips to shopping centers and stores was published in 1963. He was then hired by UCLA as an academic and very quickly promoted— I think in record time— to tenure partly because of his early accomplishments. And he later moved to the University of Texas at Austin where he spent many years until his retirement.

The Huff model is theoretically elegant and remarkably powerful. It was popular at the time it was published, but over the next 15 to 20 years, its popularity only grew. The raw materials of the model are the attractiveness of the destinations and the distances between the origins and the destinations. Unlike most other retail site models, Huff system models deal with the dollar flows and ultimately the sales of all the stores or shopping centers in a market. They cleverly capture the complex competitive effects and inter-dependences of a retail landscape and they permit “what if” analyses that can be very powerful a whole range of disciplines. A significant contribution was made in a 1974 article by M. Nakanishi and L.G. Cooper, who showed how the model could be estimated with a simple least-squares linear approximation. But this contribution only increased the quality of Huff -type research by encouraging the development of more such models and more accurate ones—which only served to increase David’s fame.

Over the years, the Huff model became an important part of countless courses in marketing, economic geography, economics, retail research, urban planning and decision theory. It is uncomplicated and compelling, uses maps and diagrams, and illustrates the statistical problem of calibration clearly to most students. Simple examples can be used to master it, and students can work out sample probabilities quickly to understand how it works. His model was included in many textbooks since 1963—making David something of a rock star among students and teachers. But David never really took the fame seriously, despite the many awards he received over the years. He was much too modest for that.

Although I didn’t meet David until the 1980s, I was already well-acquainted with his Huff model as a geography student and graduate student at the University of Toronto in the late 60s and 70s. As an assistant professor at the University of Minnesota, I relied on the Huff model for a journey-to-work model. And when I joined the Compusearch geodemographics data and research firm in Toronto in 1980, I headed a team that developed over several years a Huff model as the basis for several large bank patronage and branch success models for Canadian cities. At the time, I communicated with David on some esoteric issues, and he was always generous with his time and expertise. My team and I continued to use these models that David helped us refine for many other retail location problems for many years.

I got to know David even more in the late 1980s when I was teaching at Ryerson University and attending the annual Applied Geography Conferences. I often got together with David and other colleagues during the conference and, in the evening, we would all meet for drinks, analytics war stories and friendly ribbing. David was often the center of attention and the life of the party. He always had something to say about one’s tie—or lack thereof. He would tell interesting stories of his consulting practice but, more important, he wanted to hear from us about how we applied our Huff-based models, especially in large and complex settings.

At one Applied Geography Conference in Toronto in the early 90’s, I was asked to present a keynote presentation comparing buying and store preferences in Canada and the U.S. For research, I used Compusearch’s Lifestyles cluster system, and the popular PRIZM cluster system in the U.S. David and I worked together on selecting material for the presentation, which proved to be one of the funniest ever. We described America’s lower classes and Newfoundlander fisherman both eating Twinkies. We profiled American and Canadian Coke drinkers, ice tea drinkers and bourbon drinkers. We analyzed the owners of pick-up trucks versus full-size sedans. Although all of the research was genuine, we selected the most titillating parts to tell a true but often funny story, and that’s when I really got to appreciate David’s sense of humor. After the presentation, David visited Compusearch’s Toronto offices where we showed him a number of our projects. He was quite impressed to see a spatial analysis company in action along with our colourful Huff probability maps that he would go on to use in his own presentations.

At this particular conference, I remember a group of us decided to walk downtown for drinks. At the corner of Yonge Street and Dundas, we were approached by a scruffy man wearing a classic large trench coat. He opened the coat to reveal hundreds of watches. David was transfixed by a large supposedly Rolex watch which he was considering buying for about $100. Several of us warned him that it probably lacked any inner workings and he should just walk away. But David wasn’t so sure, insisting that Canadians are all honest and truthful. But we finally convinced him to take a pass. For years, David would claim that I deprived him of his only chance to buy a Rolex watch at a price that he could afford on his modest salary. Eventually, a co-worker found a knockoff Rolex watch on a website for less than $100, and we sent it to David. He loved the watch and thanked me for reversing my mistake at my expense.

Years later, David would get the last laugh when he invited me to present a joint paper with him on Huff-type site models at a conference in Chicago in 2005. By way of introducing me, he told a slightly revised version of the story of the watch, describing me as an underemployed modeler who was selling those watches in Toronto—and whom he had “discovered” on a street corner. I am not sure whether David’s knockoff Rolex actually kept good time, though I suspect it is in his top drawer still, working or not.

David Huff was a great friend, a great person to talk with and a great colleague with whom to share research results. He was always fun even when he asked the sometimes painful questions about what assumptions we had made or pointed out what flaws may be in our data . He invariably provided advice and suggestions without hesitation and without the need to claim credit.

David came up with an amazing model at the best possible time. He made a huge contribution to the business of building site models for retail and banking systems. And his Huff model has been the basis for hundreds of models that my team has built over the last 25 years—significantly increasing the quality of our shopper and dollar flow projections as well as sales estimates and projections. At Environics Analytics, the marketing analytics firm where I am honored to serve as Chief Methodologist, we’ve used Huff models since the company’s founding in 2003. And I know we will continue to use Huff-based models in the future; that is David’s legacy and our tribute to his enormous and enduring contribution. David will be missed by many, including his colleagues and friends, professionals at companies like Environics Analytics and the whole applied geography industry. And he will be missed by me.

Back to top