Why Supply and Demand Are Hard to Measure

Over at Economix, Catherine Rampell asks, “Does lowering the price of broadband increase its use?”

This provides a useful teaching moment. She collected data on broadband prices and adoption rates in different countries; and by linking lower prices with more broadband adoption, she’s trying to figure out the demand curve.

Unfortunately though, empirical economics isn’t that simple. Imagine instead that a supply-obsessed economist were interested in asking “Does increasing use of broadband raise its price?” Similar logic might lead him to examine data on broadband prices and adoption rates — yes, the same data — but he would expect to see more broadband correlated with higher prices because the supply curve is upward sloping.

So does a graph of broadband prices and quantities in different countries tell us about the supply curve or the demand curve? Unfortunately, it’s a mishmash. Let me explain. Prices and quantities are determined by both supply and demand. If both curves were the same in every country, broadband prices and use would be the same in every country. However that’s not what we observe. Prices and quantities differ across countries. But is this because their supply curves differ or their demand curves differ?

If the supply curve differs across countries — perhaps because it is more costly to lay cable in some places — then some countries will be on the leftmost part of the demand curve (high price, low quantity), and others will be on the rightmost part of the demand curve (low price, high quantity), with some in between. This is case 1 in my chart below; in this example, the cross-country data would be where the black crosses are. This must be what Rampell has in mind when she says “As Econ 101 would predict, the two measures are related: prices go down, subscription rates go up.”


But this isn’t all that Econ 101 suggests. What about case 2, where the demand curve varies across countries? People in rich countries like the U.S., Western Europe, and Scandinavia are probably willing to pay more for broadband access than people from Poland, Turkey, Mexico, or other poorer countries. My graph shows that poor countries will be on the leftmost part of the supply curve (low price and quantity), and rich countries will be on the right side (high price and quantity), with some in between. This is not just a story just about G.D.P. by the way — there are many other reasons demand may differ across countries. But whatever the reason, the black crosses representing the data would suggest that high prices are associated with high quantities, even though all demand curves are clearly downward-sloping.

So we have two cases, both of which have downward-sloping demand curves, but in one, the quantity of broadband subscribers is low in countries with high broadband prices; and in the alternative case, broadband subscriptions are high in countries with high broadband prices. In reality, the world is a mixture of both cases. The conclusion from this little example: when you plot real-world price and quantity data, you don’t learn the slope of the demand curve (unless you are strictly in case 1), and you don’t learn the slope of the supply curve (unless you are strictly in case 2). Instead you learn a combination of the slope of both demand and supply, and the extent to which variation is driven by these two forces.

The data that Rampell compiled are shown in the graph below. At first blush it looks like a mystery, as broadband price and quantity don’t look to be closely related, leading her to ask: “So what gives? Why isn’t there a stronger relationship between price and use?”

What gives is that the real world is serving up healthy doses of both case 1 and case 2: both supply and demand curves are different around the world. Consequently, we see low broadband prices yielding high adoption rates in case-1 countries (as Rampell expected) and low adoption rates in case-2 countries (check out all the poorer countries below her regression line, below). And the reverse pattern holds for high broadband prices. That is, you can’t infer supply or demand curves from simply looking at price and quantity data. For the technically minded, this is called the identification problem, and it is why econometrics is so darn difficult. This problem — the bane of most economists’ lives — arises precisely because prices and quantities are determined by both blades of the supply and demand scissors.

Armed with a little bit of economic theory, the poor fit between price and quantity isn’t such a mystery after all.


Robot Mistake

As a non-economist is there any way to weight the fact that for a 'rich' country broadband is more a necessity while in a poor country it may be a luxury to have broadband.

Additionally, income skews the opprotunity cost of broadband ownership.


@ robot mistake

the difference between a necessity and a luxury would be picked up by the demand curve. That is, where it is a necessity the demand for it will be higher.


Is there any factoring in of Marketing in these curves? Because if marketing has anything to do with driving demand, then it's even harder to measure supply and demand.

Michael F. Martin

Thanks for the teaching. If supply and demand curves shift around in different geographical regions, or in the same geographical regions in different periods of time, then why do we get to assume that economic equilibrium is well-defined? Whether you assume the shifts occur because of how preferences are distributed or because preferences change, it seems that a basic stability criterion fails -- there's no guarantee that exchanges are approaching an efficient allocation of resources.


I imagine that the primary cost for broadband providers is *expanding* their network, not providing access to existing customers.

Thus, countries where most providers are in the middle of expansion will probably have steeper supply curves than countries where most of the network is already built.

Perhaps the relationships would be clearer after taking into account this fact.


Why do you assume that supply and demand curves are linear? If they are higher-order, then you would have different, even opposite slopes at different points along the curve. By the same token, econometrics seems like the worst kind of "parking lot" science - let's look for the answers where our theory predicts them rather than where they are.

Ultimately, broadband in particular is subject to network effects rather than economic effects - broadband becomes worth more when more of your friends and neighbors are connected. As a result, the adoption curves are exponential.


I thought my Econ 1A professor spent a lot of time explaining the difference between demand & quantity demanded, and supply & quantity supplied. I'm glad he did because much confusion is caused by not knowing the difference.


What I don't see listed is what percentage of population has computer access/Internet in general. If 20% of a population has computers, and 15% has broadband, then there is more users for a price point, than if 80% have computer access but only 15% have broadband. Has any research been done that shows broadband adoption in relation to computer ownership and/or standard dial-up users?

Likwise some countries have greatly varing rates on broadband adoption - which simplifed example this doesn't account for. In some rural places in the US for example, you can't get broadband regardless of how much you would be willing to pay for it.

This is a problem I had with my intro to econ classes - they tried to narrow everything down to only one or two variables, when in real life its never that simple.

Bobby G


You could argue that your "network effects" are in fact economic effects; think of economies of scale on the supply side... now with broadband you just have an economies of scale on the demand side as well.

Juan Z

A better fit would probably result from examining average price within a country. I particular in urban areas there are multiple broadband sources available and probably lower prices as DSL and Cable Modems use existing physical links. In other areas of the country there may be limited broadband and only more expensive options.
My guess is both demand and supply matter. But comparing countries adds too much noise.



We don't assume that the Demand Curves or Supply Curves are linear.
However the Gauss-Markov Theorem states that if we wanted to estimate the slope of either of these curves, then the best way to estimate them is with a Linear Regression line. A linear line won't always give you the best fit, but it undoubtedly is one of the first tools in the Econometrics tool box to try. I'm assuming this is just an exploratory thought exercise of course ;)

I assume if we had more data and a clearer understanding of the relationship we could try to fit a logarithmic or exponential line to the data.

Graig Eldred

I, too, am a non-economist. I consider broadband more of a utility requiring a lot of infrastructure before it can be offered. Does the same identification problem exist with consumer commodities that are not so much like utilities? Or is Adam Smith more on the mark in the more common case?


Uruguay is a relatively poor country with relatively high internet and broadband penetration, but with extremely expensive monthly subscription prices (max. broadband is 3 MB, 1024 kb cost about us$60).

Steven F.

One look at the R^2 and it is obvious that her estimation technique (or the model itself) is flawed.

Economics is more than just running regressions. Freakonomics failed to communicate this regarding abortions vs. crime rates in the book. The actual paper, which was a much more rigorous analysis, did. I hope this post illustrates why a Ph.D. in economics has value.

Applying the basic principles of economics to a complex world problem in exchange for a simple answer is like making an omelet without butter and a pan...messy.

Kevin M. Arts

This is a little basic. I think she was asking not only why her technique wasn't picking up the expected relationship, but also why identification was proving difficult. If you look at the comments following her article, they are (for the most part) implicitly aimed at explaining variation in both supply and demand. Isn't it the point of econometrics to dig a little deeper?

Anyway, while its true that each country might have unique supply and demand curves, shouldn't broadly similar countries have broadly similar curves. That is to say, why is Norway so different from Sweden and Finland? Or France, Germany, and Austria? Is it due to idiosyncratic business environments, regulations, etc.? Maybe different subsidies? How would you propose getting around these confounding factors?

Kevin M. Arts


Upon a second reading, her model does only include price and quantity. If she wants to find the structural demand parameters, she should include an exogenous variable that affects the supply function, but not the demand function. Supply subsidies might be a quantifiable variable. Any others?


One of the things I have done is compare the broadband numbers to household usage of PC's as the OECD says this is one of the main indicators for broadband penetration. I have put these in a Google Motion Chart. Have a look here


But there is much more to these numbers than meets the eye. Having an unbundled local loop is often positive to broadband penetration. The OECD is the source on what works.


Thanks for the lesson indeed. To make matters somewhat more complicated: is not there an inherent assumption here that we are dealing with perfect, full competitive markets. That seems hardly realistic as the telco market is heavily (price) regulated even in so called liberated markets (e.g. the Netherlands). Furthermore supply may be semi monoploistic (in NL at my home address I can choose from exactly 2 broadband suppliers)


Juan Z has said it correctly, but also of concern, as someone stated, is the broadband penetration/CPU usage.

All in all, this is a too over-simplified model - if there is such a thing.


So.... What happens to the regression when you drop the outlier, the Slovak Republic? I don't have the data but to my naked eyeballs, the scatterplot looks like there is very little correlation.