I am finally getting around to writing about the third paper in my MIT PhD thesis. The first paper was an empirical analysis of the impact of information technology on firm size. The second paper provided a model for how improvements in information technology can result in hybrid organizations. The third paper is both the most ambitious and also the most theoretical — it attempts to provide a general framework for analyzing communication in organizations. In rereading it I found a fair number of errors in the conversion from a paper copy to HTML which I have not yet had the time to fix, so apologies up front to anyone attempting to read it.
The motivation for the paper was that with changes in technology we have seen substantial changes in the nature of communication — what is communicated and how it is communicated. For instance, work by Joanne Yates documents the standardization of business communication around the time of the industrial revolution. She treats this standardization as an important innovation in its own right that enabled commerce. My paper was an attempt to show that this kind of standardization can be the natural outcome of a process of designing the optimal communication protocol in response to changing technology. Or put differently, the change in the communication technology (the rise of the telegraph) is the exogenous factor and the standardization of business communication is endogenous.
The basic structure of my model is that one party observes a state of the world and another party needs to take an action which has different benefits depending on the state. Without cost of communication it would always be optimal for the observer to transmit the exact state of the world so that the actor can choose the optimal action (with full information). In my model I introduce three types of communication costs: designing a protocol, encoding and decoding the state of the world, and transmitting the messages. The optimal choice of protocol then balances the benefits of being able to choose the action more precisely with the sum of these three costs.
The paper is divided into two parts. First a fairly general analysis of the properties of this kind of model. Second three more concrete examples: management by exception, standardization of communication, and centralization versus decentralization of decision making. In retrospect I wish I had stronger math skills when I was writing this or maybe sought out the help of a mathematician. I am pretty sure that someone more on top of their math could come up with more powerful results for the general case (part 1 of the paper). Instead, I wound up having to make fairly ad hoc and restrictive assumptions in part 2 about the structure of the state spaces and benefit functions in my analyses to get some results. This seems like an opportunity for someone who is currently in graduate school and is more mathematically capable!
This is the second post in my little project of self publishing my PhD Thesis. Last week I introduced an econometric analysis of the impact of Information Technology (IT) firm size. This time it is a more theoretical paper that presents a model for examining the impact of different types of IT on the structure of organizations. In particular, it compares having no IT, with centralized IT (think mainframes) and networked IT (think Internet and Intranet).
I was particularly happy with how this paper came out because it addresses a fundamental problem in economics: the organization of economic activity in firms versus markets. In re-reading the paper I found that I don’t actually bring this point out very well at all because I focused the paper too much on different organizational forms (inside of firms). So I will try to do a better job in this little recap / introduction.
Ever since Coase published his seminal essay on The Nature of the Firm, there has been a long running inquiry into a fundamental question of economics: why are some activities carried out in the market and others inside of firms? Coase and subsequent writers focused on the idea of differing transaction costs, but the precise mechanism by which transaction costs would be different in a market versus inside firms were hard to pin down. That changed with the work on principal agent problems and incentives. Tons of different economists contributed to this. I was fortunate to have one of them, Bengt Holmström, as one of my thesis advisors.
One of the key insights coming out of Bengt’s and others’ work is that firms exist to reduce incentives. Why would you want to reduce incentives? In order to get better coordination. If you pay people a flat wage then you can direct what problems you want them to work on and how you want them to work together on those problems. In fact, much of what companies do in HR and compensation, such as reviews, goals, options, bonuses, etc. is aimed at restoring some additional motivation in the face of much reduced incentives. Effectively you can think of this issue as a coordination - initiative tradeoff frontier. You can get more coordination inside of a firm than in the market by reducing individual initiative.
What my paper does is explore the shape of the coordination - initiative frontier based on different types of information technology. I distinguish between three different scenarios: no IT, centralized IT and networked IT. I examine how these three different scenarios would play themselves out in the absence of incentive problems, i.e. when everybody does the “right” thing to maximize joint production. That analysis provides a baseline for looking at the incentive case. In the incentive case people look to their own benefit first and thus exert less effort if their incentives are muted.
The key findings are summarize in the following diagram
The diagram shows the location and shape of the coordination - initiative frontier. The first key finding is that having information is a good thing as the frontier for both IT cases dominates the one without IT. The second key finding is that that networked IT dominates central IT case: it is possible to achieve full coordination at a much higher initiative level than with centralized IT.
In the paper I interpret this result along different organizational forms inside the firm. But it would be just as correct (and looking back at it more relevant) to classify this as the historical change that we have experienced from mostly individuals in the market (agriculture, crafts, trading), to having large hierarchical firms (industrial society), to replacing those hierarchies with networks (now). That of course is a very nice fit with what I and others have been arguing in the Peer Progressive agenda.
As previously announced, I have self-published my MIT PhD thesis. Today’s post is about the first paper, which is titled “Information Technology and Firm Size” which — as the rather plain title suggests — examines statistical evidence of the impact of the increased use of IT on firm size. Before starting any analysis of data it helps to have a hypothesis and so the question is should we expect to see larger or smaller firms? Unfortunately, there are many different ways that computers might impact firm size and some of those work in opposing directions! For instance, reduced coordination cost might lead to smaller firms but better use of information assets might result in larger ones.
The paper sets out all the forces I could think of and discusses some of their theoretical background. It then proceeds to compare the manufacturing and retail/services sectors for which the relative importance of the different forces shakes out differently. This is really the key idea of the paper. Because Census data on number or establishments and establishment size (establishment is roughly equal to a location) is available by industry it should be possible to test whether the impact fits with the hypothesis.
What I remember the most from working on this paper is how painful it was to assemble the data (which makes it all the more annoying that many moves later I seem to have lost it entirely). In particular, the Census bureau has a habit of changing definitions of industries and even measurements over time. Some of that is of course unavoidable as the economy changes, some of it seemed distinctly arbitrary. Also, my thesis predates much of this information being available for download and I rekeyed a lot of it from printed Census reports.
In the end though I was very happy with the results. Here is the summary of my findings for manufacturing:
For the manufacturing sector, the dominant effects appear to be the increased flexibility of physical assets, the heightened importance of skilled human assets, and the reduced coordination cost. As discussed (…), all of these effects favor smaller firms. Both in the correlation and the regression analyses, the coefficients point generally in the hypothesized direction. Information assets appear to have the hypothesized effect of leading to larger firms, but for manufacturing they are outweighed by the effects of physical assets and human assets.
And here by contrast is the summary for retail and services:
For the retail and service sectors, the dominant effect instead appears to be the increased importance of information assets which results in larger firms. The influence of physical assets seems to be insignificant for both the retail and service sectors. The results for human assets were somewhat inconclusive, with higher skill levels associated with larger firms in retail and smaller firms in services.
What’s particularly comforting is that these trends seem to have continued since. It is somewhat shocking to see that my data ends in 1992 (!) because the Census bureau used to be (and maybe still is) several years behind in publishing the data and I started working on this in 1996. Since then we have seen some massive growth in retailers (just think of the huge drug store chains) and also financial services firms. By contrast if anything it would seem that at least for US manufacturing we are likely to have even more small firms.
It would be terrific if someone were to update this analysis. For instance, it would be interesting to check if we are seeing an increasingly bi-modal size distribution. For instance, in banking we now have a few mega banks but it now also possible to get a “bank in box” from service providers and be up and running as a new bank with very little effort.
In 1999, which now feels like a very long time ago, I finished my PhD thesis at MIT. I had started working on it following my general exams in 1996 but promptly got very busy co-founding a company (the now defunct W3Health — a story for another day). Running a startup during the week, working on a dissertation on weekends and commuting between Boston and New York is not something I would recommend to anybody. In fact, the acknowledgments to my thesis conclude with:
Most of all, I would like to thank my wife Susan Danziger who is an inspiration for everything I do and whose patience finally ran out and made me finish this thesis.
There were many times when I thought I should simply call it quits on the thesis, but the German in me couldn’t let go. And thanks to Susan giving me a big push it got done. Thanks again!
In retrospect I have been happy to have finished my thesis, which takes the form of three separate papers. My only regret has been that I didn’t have the time to polish the papers sufficiently to submit them for publication. As a result they have been rotting in the archives at MIT. A while back I finally decided to get a scan and then paid a firm to convert them to HTML.
So now many years later I have published my thesis online. Over the next few weeks I will re-read each paper and publish a brief summary as a blog post. If you are curious you can go ahead and start reading now.
P.S. I have already discovered and fixed some conversion errors but surely there are more.