Did Workers Pay for the Expansion of Products Liability Law?

May 26, 2004

Unedited transcript prepared from a tape recording

Proceedings:
MR. KLICK:  [In progress] -- which is run by myself, Michael Greve and Richard Epstein.  We try to highlight both empirical research on liability issues and institutional research, and this more or less falls in the empirical category.

It's quite an interesting paper today.  It's by Alexander Volokh, who is just recently graduated from both Harvard Law School and Harvard's Economics graduate program, will be clerking quite soon for Judge Kozinski in California.  The paper is entitled:  "The Products Liability Revolution and Labor Markets."

So we're going to start with Sasha presenting his paper, and then we have two discussants.  We will begin with AEI's own John Lott, who has written on related issues and who is cited in Sasha's paper.  Then secondly, we will move to Martin Grace, who is a professor of risk management and insurance and legal studies at Georgia State University.

So without further ado, Sasha?

MR. VOLOKH:  My paper is called "The Products Liability Revolution and Labor Markets."

So basically here's what my paper is about.  In the '60s and '70s, as we know, there was a substantial expansion in products liability as a matter of state law, and I was curious what effect that would have on the wages of people who make these products that are the subject of products liability and people who use them on the job.

Essentially, we would expect two separate effects as a result of the expansion of products liability.  Basically, the expansion of products liability--I'm going to get more specific on what exactly that means, what doctrines I'm talking about specifically in products liability--but for right now let's just say expansion of products liability, meaning that plaintiffs who suffer on-the-job injuries are more likely to get more money for whatever injuries they suffer on the job as a result of the characteristics of products.

In any case, if you're in an industry that produces dangerous products, like if you are making construction equipment that might injure somebody at some future time who's using it, then when products liability expands, that makes your employer less productive.  It reduces the marginal product of labor, and you might expect that as a labor demand effect the wage of workers in industries that produce dangerous products would fall.

In general this is ambiguous because you might also expect it to be taken out on the employment in dangerous industries.  So you might imagine in a perfectly competitive industry you might have an effect where you only have an employment drop and no effect on wages at all.

So depending on the characteristics of the industry you might expect to see some drop in the wage of workers who work in industries that produce dangerous products.  The other effect is labor supply side effect.  Essentially it works like this:  If you're a worker in an industry that uses dangerous products, then let's say in the days before the expansion of products liability, the only person you could sue conceivably is your employer, but in reality you can't sue your employer because you have Worker's Comp, and Worker's Comp precludes remedies against the employer.  So if you get injured you get your Worker's Comp payment and that's it.

But now suppose that I'm working at AEI and my hand gets shredded by a Xerox copier, now thanks to the expansion of products liability, I can sue the distant manufacturer, Xerox.  So from my perspective, this is an extra perk of the job.  It's an opportunity to participate in an extra lottery, and the theory of equalizing wage differentials predicts that when you have an extra perk of the job, whatever it is, it would tend also to lower wages, because if it didn't more people would want to work in the job because there's this extra perk.

One of the assumptions here is that Xerox, for instance, let's say we're in D.C. and this is done as a matter of D.C. law.  Conceivably, Xerox could say, oh, you're buying a Xerox copier in D.C., and pass the increase in cost onto AEI, so an assumption that I'm making is that these products are sold on a national market so that when the state level change in tort law happens the product manufacturer can't just smoothly pass that all on to the employer.

So in any case, so put all that together and you have two different effects which you could predict would reduce wages for workers in these different industries.  One again is the labor demand effect because if you are in an industry that produces dangerous products, your employer is less productive, and a supply side effect, if you use dangerous products then you get to participate in this extra lottery, and this is a lottery which is more valuable for industries that are more dangerous.

My bottom line--I'm just going to flag it ahead of time, but we'll get to it later--is that the doctrine of competitive negligence shows up as having a negative effect, reducing wages by about 4 percent.  And there are further complications, but we'll get to those in due course.

So product liability of course is not a unitary animal.  It is made up of a lot of different doctrines, and I focused on these five doctrines because gathering the data here meant reading a lot of cases from a lot of jurisdictions and trying to locate the case in whatever, 1968, where they say, "Therefore, we hereby adopt strict liability as the law of this state."  And if I can find that, then I can code 1968 as the year that that state adopted that doctrine.  So I focused on five doctrines.

One is the adoption of strict liability for manufacturing defects.  That's usually associated with a restatement, Section 402(a).  Number two, whether--when you're litigating over a design effect, whether the manufacturer is off the hook if you can show that even though there was something hazardous, it was fully known by the consumer and perfectly accorded with consumer expectations.  Originally that would get the manufacturer off the hook, but later it came to not get the manufacturer off the hook, so that was component two.  Three, whether the state has shifted from contributory negligence, where if you're a negligent plaintiff you get zero, to comparative negligence, where if you're a negligent plaintiff and you can show that the defendant was also negligent, then if you're half and half negligent, you get half.  Number four, whether punitive damages are available in a products liability action.  Number five, whether having an inadequate hazard warning qualifies as a species of design defect.

So that's five things which their overall effect is to--I think it's fair to say, broadly speaking, their overall effect is to increase the likelihood that a plaintiff would prevail at trial and also the amount of money that he would collect conditional on prevailing at trial.

Now, the question is, what sort of likely effect should we expect?  Is this something very important or should we expect something really tiny?

In my paper I do some back-of-the-envelope calculations, basically comparing how much the Worker's Comp system pays out to how much the tort system pays out, and after doing these back-of-the-envelope calculations, essentially the amount paid out by Worker's Comp is on the order of something like 20 billion, and actually I'll just--the Worker's Comp system--this is in the '80s--took in 22 billion and paid out 14 billion, and if you look at the tort system the general liability insurance premiums are about 20 billion, and when you take into account how many injuries take place on the job--and only about one-sixth of injuries take place on the job--but the amount of loss is much larger for those, and also the amount recovered is larger for on the job injuries than for not on the job injuries.

It turns out that on balance it's about--the amount of money that passes through Worker's Comp and that passes through the on the job products liability portion of the tort system are of roughly comparable orders of magnitude.  So if we come up with an effect for this which is comparable to the effect that we find for Worker's Comp, we shouldn't be surprised.

So some other work, Fishback and Kantor and some others have established that workers did pay for the adoption of Worker's Comp laws when you take into account how much workers could expect to get from the Worker's Comp system, you find that on average their wage dropped by about exactly that entire amount.  That doesn't mean it's a bad system.  In fact, they used that to conclude that Worker's Comp is on balance a good system, but in any case, just the positive effect is that workers did pay for the adoption of Worker's Comp laws.  Moore and Viscusi come up with an estimate of about $1,475 as a kind of a yearly wage effect of the Worker's Comp system.  Lott and Manning had a paper where they show that workers did pay for an expanded ability to sue input suppliers as related to the Borel v. Fibre-board decision on asbestos, and their estimate is roughly on the order of $1,100.

Let's go on to my empirical strategy.  Basically what I'm doing is something very simple.  My left-hand side variable is the wage, and my right-hand side variable is tort stuff, and by tort stuff I'm going to move those five doctrines that I talked about.  They're going to show up as five indicator variables.  Tortk, which is the state and Tortt which is the time.  So, for instance, we'll have punitive damages sub Alabama 1970, which will be zero if Alabama in 1970 didn't have it, and 1 if it did have it.  We're going to have the injury rate.  That's something I got from Bureau of Labor Statistics books.  We have Worker's Comp which I got from analysis of Worker's Comp books.  A bunch of dummies.  Ij is an indicator variable for the occupation.  Ik is dummy variables for the state.  It are dummy variables for the year.  So I've got fixed effects there for occupation, state and year.  The Xi is just a big vector of demographic things which I get from the current population survey, and the wage also I get from the current population survey.

So that's my first specification.  My second specification is, basically, when you look at the sign of beta1, if beta1 shows up negative, that kind of proves the thesis.  Of course we don't really have beta1.  We kind of have beta1 punitive damages, beta1 design defects, beta1 strict liability and so on.  In any case, all of those could be negative.

But as I said before, the benefit of the products liability system is higher for people who work in more dangerous industries.  So I interact the tort variables with the injury rate, and so we might expect the interaction variable on--the coefficient of the interaction effect.

Another thing, and this is a little bit mushier, but I wanted to somehow get a handle on what was the size of the demand side effect versus the supply side effect.  So recall the demand side effect is when you're producing dangerous products which might injure somebody, and the supply side effect is when you are using dangerous products for which you might get injured and then recover money.  Both of those should be expected to reduce the wage.

So what I do is I construct an indicator variable.  I call it I19/39.  SIC Codes 19 through 39 are basically manufacturing industries.  So it's industries that make stuff.  That's actually 20 through 39, but I threw in 19 also because it's firearms, so I figure that seems appropriate to throw in there.

So the idea is that in industries that produce dangerous products, you might find a larger effect, and in industries that don't produce dangerous products, where they're actually just using products, you would find a smaller effect, and by looking at the difference between the two, you can get a handle on the relative magnitude of supply side and demand side effects.

Also just for you empiricists out there, just doing this, using ordinary Lee squares, requires assuming that all the epsilons are independent and identically distributed, but in fact we think that epsilons are correlated within state, so I clustered by state to take care of that.

Here's just a little sample of what the data looks like.  So basically like California, you know, in 1963 they adopted strict liability for manufacturing defects.  New York in 1975 adopted comparative negligence, and basically this is the result of my Lexis and Westlaw legal research, reading all those cases to figure out what they did.

So here are my basic regression results.  First let's look at the first column, which is Regression No. 1.  That's the regression where all I've got is the tort variables on the right-hand side.  So essentially unless it's got stars, ignore it because it's not significant.  So the only thing that comes in at all--so first the injury row, this is a known equalizing wage differential effect.  Ever since Adam Smith pointed it out, everybody knows that if you keep other things equal, if the injury rate is higher in a profession, the wage should be higher in a profession.  So we want the injury coefficient to be positive, and indeed it is positive.  Also Worker's Comp.  We also know from various studies, including Viscusi and co-authors and Fishback and Kantor and so on, that when you have higher Worker's Comp payments, that is a perk of the job, therefore it ought to reduce wages, so it would be nice if Worker's Comp comes in with a negative sign, and indeed it does come in with a negative sign, though you can actually ignore it because it's not significant.  It could be zero, but at least it's not positive and significant.  So those are two good things.  At least the stuff comes in consistent with what we would expect on injury and Worker's Comp.

Now let's just focus on the five tort variables.  So in Regression No. 1, the only thing significant is comparative negligence, and comparative negligence ends up decreasing wage by about 4.3 percent.  In a minute I'll get to why that makes sense.  Actually, I'll say it right now.

Comparative negligence is not something that most people think of as being a highly important doctrine, but one of the aspects about competitive negligence in the way that I collected the data, I had to read these cases and figure out when the state adopted the doctrine.  Now, some doctrines are really crystal clear.  That is, they didn't have the doctrine before and then the Court says, "therefore we now adopt this doctrine," and then it's crystal clear that that is the new doctrine.  Comparative negligence is kind of like that.  The other doctrines are a lot fuzzier, and so it can be hard to figure out when exactly the doctrine is adopted.

So what that means is that I'm measuring the year of adoption with some measurement error, and when you have measurement error, that means that you have attenuation bias, that is, the coefficients that you find are smaller than they actually are in reality.  So if you find a lot of stuff which is not significant, you might expect that when you have attenuation bias, when you measure things with error.  But on the other hand, comparative negligence I measure with very little error I think, so there again, it's not surprising that that comes in more significant.

Also, comparative negligence is correlated with a lot of other stuff because whenever comparative negligence comes in, if you're a negligent plaintiff, you used to be getting zero instead of (d) your amount of damages that you would otherwise be getting.  Now you get Alphad where Alpha is your proportion of non-negligence, or let's say the defendant's proportion of negligence.  So when comparative negligence you go from zero to Alphad.  So every pre-existing doctrine that came in as a determination of (d) is now magnified.  So essentially comparative negligence magnifies every pre-existing doctrine.  So in a sense it's correlated with the introduction of other doctrines.  So that's kind of an omitted variable by us which would tend to boost this one.

So I would say that the 4.3 percent is not actually the effect of comparative negligence as such, but it's actually a more general effect of the tort system which is showing up on comparative negligence because it's more precisely measured and because it's correlated with other omitted doctrines.

Now let me just breeze through these other regressions here.  Regression No. 2 is where I interact with the injury rate.  Now here the only things that come in significant are again comparative negligence and strict liability.  Now, these numbers you can't actually interpret them off the screen because you have to take the first column plus the mean of the injury rate times the second column.  Once you do that you find that on comparative negligence you get almost the same thing again, about minus 4 percent.  So that's consistent.  On strict liability, it's a positive interaction term which is surprising, but you get approximately a zero effect over all.

The best I can do for explaining why strict liability would come in with a positive term is because when strict liability replaced negligence for manufacturing defects, it's not the way you might read it in some tort theory by Chevell [ph].  It's not that it was the strict liability system replacing an exactly equivalent negligence system.  The negligence system by that time had become so riddled by exceptions, particular sub-doctrines for particular industries, essentially when courts adopted strict liability, they themselves thought that they had been applying strict liability all the time, but under the table calling it negligence but carving out all these exceptions.

I think that what we should expect the effect of strict liability to be actually depends on a lot of very doctrine-specific things like how high the standard of care had been set for negligence and so on.  So I think that once you take that into account, it's not surprising that we get an interact effect which is slightly positive but overall kind of washes out to zero.

Finally, my Regression No. 3.  This is where I interact with the indicator variable that indicates whether you're in a manufacturing industry.  Regression No. 4 is where I interact with both and that's really the important one, but this is just here just to give you a flavor.

On this one comparative negligence comes in significant again, and here you find that essentially if you're in a non-manufacturing industry, you're like minus 3.3 percent, but if you're in a manufacturing industry you add these together and you get roughly minus 6.1 percent.  When you look at that you see that roughly speaking being in a non-manufacturing industry gives you kind of half the effect of if you were in a manufacturing industry, and that kind of tells me that the supply side effect and the demand side effect--again, this is not rigorous but kind of touchy-feely.  It kind of tells me that there are roughly similar orders of magnitude.

It turns out the supply side effect is about 40 percent of the total effect for comparative negligence, about 80 percent of the total effect for strict liability.  I would consider that in each case kind of roughly of the same order of magnitude.

I already explained why the CN effect makes sense, No. 1, attenuation bias.  I measure this thing more precisely in correlation with omitted variables.  I've explained why the strict liability effect might make sense.  It's not perfectly clear but I think that it's not on its face crazy.  And the size of the effects make sense because if you take negative 4 percent and you evaluate that at the average wage, which is a weekly wage of something like $680, you get--no, sorry--an average weekly wage of $327--this is a sample that goes from the early '60s to the late '90s, so the average weekly wage here is $327.  And so a 4 percent drop in wage comes out to a yearly effect of about $680.

If you take, for instance, let's say a 40 percent multiplier to get a supply side effect, you get to $272.  In any case, we're talking about numbers that are roughly on the order of 300, 500, 600 dollars.

Now, compare that to Viscusi and Moore, who say that the Worker's Comp system gives you effects of $1,475.  Or compare that to Lott and Manning, who come up with overall effects of on the order of something slightly more than $1,000.  So in any case this number is not crazy because it agrees with John Lott.

Let me just wrap it up.  That is the bottom line of this paper which is still a work in progress, but kind of these tentative results, is that the tort system really does have an effect.  It does reduce wages somewhat in dangerous industries that use and in industries that manufacture dangerous products.  As I said at the beginning, you would expect that some of the demand side effect would be taken out on wages and some of it would be taken out on employment.  I actually did run some regressions where the left side variable was not wage, but the number of people employed in these industries, and I did find that the interaction effect, I interacted again, the tort variables with injury because you would expect that the effect would be larger in more dangerous industries, and I did in fact find that the interaction effects were negative, so that the tort system variables do have a certain effect on employment.

Where I didn't find that was for strict liability, but again, we found a slightly positive effect for strict liability on the wage, so that's again not surprising.  The employment effect kind of winds up there, but for the other doctrines you do find some small effect on employment.

Now the question is what are the normative implications?  I personally would not draw any normative implications from this work at this stage.  At this point, if you really believe what I've said--not clear whether you should believe what I've said--but if you do, all that this has shown is that the adoption of this products liability system has decreased wages.  But let's think about what Fishback and Kantor say about Worker's Comp.  They show that Worker's Comp gives benefits to workers and also their wages drop by about an equivalent amount because their conclusion is this is great for employers because they don't actually have to pay for it on net because they pay Worker's Comp fees, but then they pay less in wages, so they benefit, and the workers benefit because they gain a benefit but they lose wages, but overall, they gain in insurance value.  They're more likely, they're more certain to get compensation in the event of an injury, and they don't have to go through all of the hassle of suing their employer on some negligence theory which might not pan out, and maybe they might get a huge amount, but maybe they might get nothing.

So in any case, their conclusion is that just because you find the wages dropped doesn't mean that this is a bad system, and so hereto, just because you find that wages drop, all this means is that to the extent that distributional concerns are motivating you in advocating what sort of tort system to have, you should take into account that the on paper distributional effect will be somewhat offset by this reduction in wage.  That is the humble normative implication of my paper.  Thanks very much.

[Applause.]

MR. KLICK:  Thank you, Sasha.

First we'll have John Lott comment, a resident scholar here at AEI.  He's sort of a widely cited and widely published scholar in the law and economics area, focusing on empirical work in this area and many other areas of law and economics.

Thanks, John.

MR. LOTT:  I'll wait a second for the lights.  Some of us aren't as young as Sasha.  Our eyes go after a while

But I want to congratulate Sasha on, I guess, successfully defending his dissertation apparently the other day, and his clerkship on the appeals court out in California for Kozinski, who I'm sure will be--well, at least we've got light now.  Thank you.

I suppose the first thing that struck me just as kind of an empirical person, I'm not sure I've seen too many studies that go and use 790,000 observations.  I've used a lot of observations in some of my studies, but 790,000, I have to ask you how big the data set was.  Of course, you don't have that many variables across, so that may help some.

More seriously, I guess where I start on a problem like this is firms could have provided this type of insurance anyway, and they didn't.  I mean there are people that were selling the products, and so to me it seems almost by definition it must be some type of net loss that was there.  Maybe to go and talk about, you know--it's true that wages falling by themselves don't imply anything bad, they also don't imply anything good necessarily, but it just seems to me we kind of have to ask why these product manufactures weren't working it out with the employers for these workers to go and offer this type of, you know, these types of suits or insurance type of policy anyway before the courts got involved.

I guess to me for not just the fact that even if we accept the wages falling we can't say whether it's good or bad.  It's also not clear to me some of the breakdowns that we have here, which I think even make it a little bit more murky.

One of the things that Sasha does is an important question, and I'm glad he deals with this, is kind of the demand side versus supply side effects for these.  The problem I have is I'm not sure what the manufacturing dummy really picks up.  I mean it could pick up changes for the desires for the work in certain industries over time.  It could pick up changes in comparative advantage in the United States to go and produce certain types of products over time.  You know, it seems to me it's just a big black box that I could go and put in things that could affect firm cost as well as the demand for workers to go and work in these particular industries.  Maybe you can explain more there, but at least it wasn't clear to me.

I think what he does is admirable and good in the sense that you have this panel data set.  We identify the changes in laws in particular years, and we see the effects, kind of a standard type of panel study here.  He talks about the error in measuring these event changes, and if it's kind of random error, just noise, how that not only makes it less statistically significant, but also biases your coefficients toward zero too.

But it seems to me that there may be something more systematic that's going on here.  And the simple reason is that firms will start to adjust before these changes actually occur.  It's not just--I suppose at the simplest level, I mean, it's costly to wait until a decision's actually been handed down, but once the suit's been filed even, first have some probability that they're going to lose the case.  Workers have some probability that they're going to win.  So you would think wages will start changing even when these things first get adopted.  You don't have to finally wait until the actual decision for the courts.

But I think it's even more than that because you start seeing other states win, and you start to say, well, it may just be a matter of time after X number of states have changed their law, soon somebody's going to go and bring a suit here, or a different governor or a different party wins, and I start seeing judges who may be more favorable to tort lawyers or something being appointed.

So we have event dates in lots of areas that we have problems with.  I suppose--I've always had a problem with finance studies that go and use court dates, and it's basically the same type of issue that's here.  Again, it's more than just being able to decide once you've heard the oral arguments, you know, who's likely to win once you've heard the questions?  This is, you know, years beforehand.  Once California does it, my probability in Texas that something like this will change has gone up some percentage points.

That leads to some other issues here.  You have a lot of data, and it seems to me there are things that you can go and do with this.  I mean you got this over time.  I'd like to see--and you're not alone in this--everybody just likes to use these dummies for whether or not a change occurs or not.  I guess for a long time I've tried to argue that just simple dummies can be very misleading.

I can give you a very trivial case that's here, and that is, let's say wages were systematically falling in these industries that you want to concentrate on prior to the change in the court, and then they were falling afterwards, and there's no change in the rate at which they were falling.  I go and put a dummy variable in for some change in law, it's going to pick that up.  It's going to make it look like there was some--I mean there is a different in averages before and after.  It's just that when I look at this, it may not be attributable to that because they're falling at the same rate over the whole period.  And a simple dummy's not going to be able to tell you anything about that.

Maybe if I believe the types of stories that I was saying about this being systematic, and I believe the types of overall results that you have here, I might not be surprised to see some type of systematic decline that's occurring over the entire sample that's going to be there.

There are other issues that one can look at in terms of geographic breakdowns, and I guess if you're really obsessed with the stuff, you can do time/geographic type breakdown.  But at least those two are the things that I would do.  I mean if--how long do we have to talk?

There are other issues that one could get into here, but I guess to me, when I finish reading a paper, I kind of like to know kind of what's the bottom policy, you know, whether it's good or bad, because the overall fact that you get a decline in wages, you got a really interesting data set and lots of important things that you did here, but it doesn't seem to have as much punch as we would like to normally have.

The other thing is--this is more trivial--the types of CPS data sets that you have has lots of control factors for wages.  You got not only things like you know whether people are white collar, blue collar, farm, you know whether they're married and whether they're full time or part time workers.  I mean those types of ratios change over time.  The number of jobs people have change over time.  And even with things like education and what-have-you, you break it down with those kind of trends, but you can go and put down individual year dummies and things like that to try--that would normally be what labor economists would do.

By year of education, you have trend and non-linear, but when you have this much data here--I mean that's the other issue here.  I know you deal with clustering and there's debates that one can have on that.  There are some people that have done other approaches to this in this room in fact here.  I'm sure Eric could probably talk to you about his particularly approach that he's used.  When you have 790,000 observations, it's almost amazing that you don't get like massively statistically significant results on everything.  We've got two different types of errors here, and we're using all this additional information to kind of reduce the one type of error, and so I was almost surprised that you didn't get more statistically significant results.  I mean I was almost shocked.  I'm trying to think, maybe Eric or somebody else who does empirical work can know how many studies that you've seen that have 790,000 observations.  But it's--

MR.          :  [Inaudible].

MR. LOTT:  I know he is.

MR.          :  [Inaudible].

MR. LOTT:  No, I understand, I understand.

MR. VOLOKH:  I actually cluster on state, which I only have 11 states.  If I clustered on, let's say, state times year, I would get more things that would be significant.  Just a counter argument is that, do you think that one state year is independent from another state year?  So I chose state, but I could have clustered up--

MR. LOTT:  No.  I mean I agree you're being conservative in that respect.

Anyway, I thought it was an interesting paper, and a lot of work's obviously gone into it.  I guess my bottom line though, even though I'm incredibly sympathetic to the conclusions that you have and the coefficients that you have on this, I would find it more believable if you could break down the data more and actually show some trends and stuff over time that might exist.

MR. KLICK:  Thanks, John.

Now we're going to move on to Martin Grace, whom I've asked to discuss the implications of Sasha's work or use Sasha's work as an illustration for how we might more profitably look at changes in liability systems.

Marty?

MR. GRACE:  I want to thank you, Jon, for inviting me up here today, and I'd like to thank you for sending me a nice paper to read.  I was very interested in it.  Jon actually heard me vent about something about tort reform, and I said something to the effect of, you know, we need more of a general equilibrium and less of a one-industry, one-market approach to how we think about tort reform, and so he basically told me to come up here and tell us about that, so that's what I'm going to do today.

I'm going to skip the first point because you heard it already, so that's very nice.  I want to say that I really enjoyed the paper because it's just interesting to see, you know, how people with different uses of their imagination can bring unrelated data sets together and come up with a whole new way of looking at a problem.  From that perspective I think it's a very interesting paper.

As I said, Jon asked me to talk about sort of the general equilibrium effects, and this paper does a first step in going beyond the one-market approach and it takes two sides of the workers demand market and the workers supply market, the supply of labor and the demand for labor.  It's kind of interesting how he teases out these different results, and I'd like to sort of talk about that and keep that in the back of our minds as we move forward to talk about other types of problems when we talk about how differences in liability laws will affect other markets, not just the one you're focusing on.

I have a very minor econometric issue.  It's very much like John Lott's.  I think that almost everything is endogenous and at lunchtime today we were talking about how do you know when you finish an empirical paper because there's always one more thing to do?  I said to Sasha I had that same problem, and one of the issues that I have with his paper--and it's a very minor one because you never know when to stop--is the Worker's Compensation benefit, even though it's the maximum one, is determined by a process outside the control of the individual person, but it's also one that's a political economic decision that's based upon some model that you can develop to predict what the outcome is going to be.  The same is true--yes, that's really going beyond the pale.

And this one--this is something that John kind of mentioned--that the decision to adopt a new liability law may also be endogenous and it may depend upon something that happened 20 years ago in another state across the country.  So if you can control for all of that I'm sure that you will have the pick of any type of job you want.

Now, some of the implications.  Tort costs have to be paid by somebody.  They are not magically distributed to people outside the United States, but they are borne by people in the United States.  This is a graph you see typically in the insurance industry when they focus on tort costs, but the bottom line there is kind of a normalized GDP growth, and the top red line is the growth of tort costs, and you can see that since the--I think that's 1950 down there at the bottom--that we've had quite a divergence over the last 40 or 50 years.

A lot of the literature on liability economics, liability law and economics, focuses on just one thing.  Here's just a list of things.  I just did a very cursory look.  There are mentions of the price of liability insurance, how much it goes into the price of a token on the Philadelphia subway system.  20 to 30 percent of the price of a ladder is tied up in liability insurance or expected cost.  Stock prices are affected by liability issues.  Organization and structure of the liability insurance industry of course is affected by liability rules.  R&D and innovation, frequency and severity of losses, Workers Compensation and wages, and physicians' decisions to leave or enter a market, all these things are kind of like single shot examinations of the effect of liability in one area.

I want to push this a little bit further and it's really difficult.  Jon, when I mentioned this to him about six months ago, he said, "Oh, it's too hard."  And I thought, "Oh, no, the tax people do this all the time."  I started reading what the tax people do, and it's really hard for them too.  But one of the approaches I thought we might do is push the general equilibrium approach, but maybe start in markets that don't have a lot of bleed-over effects.

I was at some hearing this Christmastime involving rural hospitals.  They wanted really to change the tort regime in Georgia because no doctors wanted to work in the rural areas if they had to pay essentially big city malpractice premiums.  So the hospitals can't afford to pay the premiums.  Not only that, the hospitals can't afford actually to have these doctors operate in their hospitals any more, whether they're surgeons or whether they're just clinicians.  So the hospitals cut services.  The doctors leave because they don't have enough patients, first of all, and then there's no place to see the patients.  Rural income decreases.  The rural area can't attract new employers because the new employers demand medical services, and if you have a hospital that is really just a clinic, you'll put your new business someplace else.  This is actually important in Atlanta because nobody wants any more new business right in Atlanta because there is too many cars.  So if you can think of a way of having businesses and doctors in Atlanta without cars, then we're all in favor of it, but we'd like to have you 30 miles away in some other area helping the rural economy.

So this general equilibrium effect basically has effects on residents, the current ones. and the future ones, there won't be any.  There will be effects on local business, the current ones, and there won't be any new ones.  It will have effects on the local tax base and the person least able to bear this risk pays the highest burden, so we're left kind of with one person with the light switch leaving, and that's the person that's going to bear the most of this tort liability.

Now, there have been some people thinking about this.  Kip Viscusi is one of them thinking about what sort of the public finance of tort liability costs actually are.  He's come up with some things that make a lot of sense.  First of all, if damages are kind of a random event, then they act as a tax rather than a deterrent.  So if we have underlying law that is noisy, then people really do not know what the standard of care actually is.  They--

MR.          :  A random event that they can't affect.

MR. GRACE:  Correct.  So therefore this is a tax rather than a way of increasing safety.  His work, Viscusi's work, has basically called into question the link between liability rule, safety expenditures and R&D on innovating around to improve safety in high liability industries.  So if damages are supposed to punish, and they don't, because they're kind of noisy, then the liability tax, if you will, gets passed on to either the customers, the investors or the workers.  So this kind of a problem that I think really needs to be pushed further.

We have some other evidence--just to show you the size of this issue--and I'm sure you're all convinced, because you're here, that it's a pretty big problem.  But Tillinghast, which is an insurance industry consulting brokerage business, does a survey every other year or so on the tort system costs, and before 9/11 they were $180 billion a year, and I think that had actually come down, but after 9/11 they were up to $232 billion a year.  These costs are really just insurance costs, things that are relatively direct that people can count.  They don't include the accident avoidance cost or the increased cost of defensive medicine or extra tests and things like that in products development.  They don't include products that would help in the market, and they don't include the loss of future innovation because we decided to stop something today.

If you look at the cost, essentially we have some 46 percent of each dollar of our tort cost goes to the victim.  There's actually a large debate that's occurring right now whether the non-economic losses which are things that cover pain and suffering, whether they actually should be given to injured victims, and some economists tend to say that those are just a lottery and a pure tax.  Others say, no, there is some science behind how they're awarded.  But even if we say that 46 percent goes to the claimant, the rest of it goes to attorneys' fees and administration.

If we look at the Workers Comp system it's a lot different.  This is a much different system.  You really can't influence the outcome very much, but the administrative costs are only 22 percent.  So if you use 22 percent as a conservative guideline, then a tort system is very expensive, as we all know.  So if this excess cost of the tort system--you know, maybe it's almost 80 percent--is assigned randomly, then somebody bears it.

The CEA in 2002--and I couldn't find who actually wrote this because it was unattributed--but they attempted to get a rough--

MR.          :  [Inaudible].

MR. GRACE:  I know, but I asked if they knew the person who I suspected of writing it, and he said he didn't.  So it's sort of forgotten.

[Tape change.]

MR. GRACE:  And there is a big debate in the corporate income tax incidence literature like who actually bears this tax.  So what they did is they covered all the bases.  They had essentially a presumption that the consumers paid the tax.  They had a presumption that the workers paid the tax, and they had a presumption that the tax was born by the investors.  Then they had at the very bottom, which you probably can't see very well, they had what if we had a mixture, which is probably more realistic.  If you look at these numbers for kind of a low, medium and high case, the 187 billion was kind of their low estimate of the actual excess burden, and the 230 was their highest, you see numbers that actually look kind of like Sasha's.  They're in the same order of magnitude in many respects.  So not only do you have numbers that are like John Lott's but you have numbers that are like the United States Government.  So you should feel good.

In sum, I think this is a great first step because we're kind of looking at more than one market.  We're looking at two markets.  Maybe we could look at three or four or five.  It's very difficult to do, but it's kind of the way we should probably think about these things.  Like a good economist, I always say more work is needed.

Thank you very much.

MR. KLICK:  Thanks, Marty.

Unless Sasha has any burning desire to address a specific issue from the commenter, I'll move to floor questions, and I'll start off with my own which is a little bit related to something to something that John said.

I'm wondering did you look at sort of the speed of adjustment, where instead of looking at a dummy variable for a legal change, how about sort of a trended variable, because presumably it does take some time for firms to adjust wages.  What I'm particularly interested is there should be some differential effects for industries or states or industries within states that are particularly unionized, right?  And I wonder if that doesn't sort of disproportionately affect your attempt to look at manufacturing industries and things like that?

So if you could discuss a little bit about some of the robustness checks that you did along those lines.

MR. VOLOKH:  Well, this is still work in progress, so I haven't done much robustness checking along those lines.  The reason that I chose the manufacturing dummies that I did--and possibly this wasn't a very good reason, I don't know--I wanted to somehow be able to distinguish between industries that make dangerous products and industries that--that is industries that are more making than using, and industries that are more using than making because that was the biggest, one of the big concerns that people had in an earlier version where I didn't do that.

So the reason that I chose those SIC codes is just because when you look down the list of SIC codes it's like, it reads like a list of people who make the products that show up in the tort cases.

MR.          :  How is that distributed with regard to how they use it?  I mean it could be--

MR. VOLOKH:  Right, exactly.  So it could be that the companies that make the--

MR.          :  [Inaudible].

MR. VOLOKH:  Yeah.  Like they use like if you're making a crane that might topple over on a construction site and kill somebody over an SIC industry, construction industry, but it could be that you're using various drill presses and whatnot to make it.  So what I really want is a kind of index of, a using versus making index.  So that's the only thing that I was trying to approximate with this manufacturing dummy, which I agree is very imperfect and it doesn't really--like I don't have strong confidence in my precise breakdown of supply side versus demand side, and if you have idea on what would be a better indicator to use, like if there actually is some data source that I've overlooked, where I can actually find out maybe if it's known what proportion of injuries in what industry are due to products, for example, or something like that.  I would really welcome that.

So in any case, that's all that was going through my head when I chose the manufacturing dummy and that's why I didn't do those robustness checks because that wasn't really my motivation in choosing that.

MR. KLICK:  We'll take questions from the floor.  Please give your name and your affiliation before your question.

QUESTION:  Hi.  I'm Ted Gare [ph] with the Council of Economic Advisers.  I also don't know who did that study.

Two quick questions.  As both discussants mentioned, it seems like so much depends on the exogeneity of your variables of interest, so I was just wondering--you didn't have much time to go through them--I was wondering if you can just expound on it a little bit.  Are these court cases out of the blue?  Are these new laws?  Are these court cases overturning recent laws?  I would think of the three, the third would be the most exogenous credibly if it's some sort of state constitutional challenge that suddenly overturns a recent law, that that would give you more kind of believability on exogeneity.

My second question goes back also to what both the discussants talked about which is what normative implications we can make.  I'm no lawyer, so I don't know the legal history, but it seems like the efficient system would be a means of kind of having heterogenous contracts that each of your employers, or across industries at least, where some you can give certain kind of product liability protection, others you can give the wage compensation and whatnot.  And I would imagine that's never been legal to have such heterogeneity.  I don't know and I'd be curious to know if it was.  And absent that, then we are looking at a question where you have kind of two state-imposed regimes, one where certain product liability rules are in place and other uniformly where they're not, and then it becomes ambiguous as to which gives you higher net benefit.  I'm curious to know your opinion.

MR. VOLOKH:  Those are both very interesting questions, and I have something to say on each of them.

On the exogeneity, basically, assuming these court cases are exogenous is something I kind of had to do because it's not clear to me what I would use to instrument for when a state would change its law.  It's true that in general the ideas that the tort system--let me start here.

The whole products liability revolution came from a development in the legal academy where all of these legal scholars--well, Prosser [ph] was the culmination of this; there were various writers before him, who he absorbed--came to endorse this theory of enterprise liability, kind of resting on the basis that individuals, individual consumers, individual workers are powerless to affect the safety of products.  They don't have as much information as these large corporations.  They can't bargain with these large corporations over safety expenditures, so that the most efficient thing is to put all of the costs onto the manufacturer, and also similar ideas show up in Calibrese's [ph] Least Cost Avoider thing.  So all of these ideas really, I think they developed very much in the ivory tower.  So if it's something that develops in the ivory tower, I'm kind of comfortable saying that it's exogenous.

[Laughter.]

MR. VOLOKH:  The problem is that jumping out of the ivory tower into state courthouses is not exogenous because you must have some reason why some--

MR.          :  [Inaudible].

MR. VOLOKH:  That's right, that's right.  And why is it that Texas jumps in 1973 and New Jersey jumps in 1977?  There must be something that's making it happen.  I just didn't do that because it was kind of difficult, but also, one thing that I learned in my public economics course is that a lot of public economists take this view about state and local innovations or state and local policy, is that if you have a lot of very rational people running the show you might be afraid that they are taking steps for a reason and then you have to model that reason.  Whereas, for a lot of things at the state and local level, things kind of happen randomly, and this may not be good from the perspective of someone who has to live under the policies, but it is good from the perspective of the statistician who's looking for exogenous sources of variation.

In any case, the reality is somewhere in between, and I just assumed exogeneity because we just don't have a very good model of what it is that is motivating judges when they go and decide to make the law X.

MR.          :  [Inaudible].

MR. VOLOKH:  Right.  I was going to get to that.  Most of these things are purely judge made law being changed by judges into other judge made law.  Like the standard has always been contributory negligence ever since the case of A versus B in 1825, and now under the pressure of this case, C versus D, this judge decides to change the rule.

Now, sometimes it's by statute.  I also similarly assume that the statutes are exogenous, even though for the same reasons there, everything is endogenous.  So mostly it's just judge made law stuff.

Let me just get to your second part which was about the normative implications.  In general I think it's--in general I'm sympathetic to this idea that, as John said, here is this service which firms could have provided before but they weren't doing it, so there must be some net loss, or as you were saying, that maybe what we would like is some sort of voluntary program where some firms would offer some benefits to some employees and other firms--other employees would be different.  Two things.

One--and Fishback and Kantor deal with this in their paper on Workers Comp.  They say that the reason that--well, they give their whole story about how it's a win for firms and it's a win for employees, so why weren't they doing it before?  The reason is because the main benefit of Workers Comp is that it preempts liability, it preempt employer liability, so the employee has to agree, make a binding contract not to sue, but those were always considered contrary to public policy and never considered enforceable.  So that that sort of statutory change was required to get around, again, this judge-made principle that certain contracts, like contracts not to use the legal system, are unenforceable on grounds of public policy.

Another thing, which even if you didn't have that, I don't want to say as a blanket statement that if firms are not providing something it must be because it's not worth providing because in these sorts of insurance like things, there's a lot of unobservability of the characteristics of the employees so you have potential problems of adverse selection.   So there's just a possibility that it's a market which wouldn't exist which would unravel if it were just left to--or even if it wouldn't unravel, it would just exist with big inefficiencies.  It wouldn't--and so that actually it could be more efficient to have a mandatory insurance system.  I don't want to say that that's true as a general matter.  It's just that in this paper I didn't want to make the plunge and actually draw a strong normative implication.  That's all.

MR.          :  It's been a while since I've read -- [Technical interruption.]

MR. VOLOKH:  Right.  And this is something where it was crucial for me to have the nationwide versus state component.  If we were all a unitary state, then you wouldn't have this effect because the whole idea of equalizing wage differentials is if you get to sue your employer, or let's say the law says the employer has to give you $10 or something like that as an outside of wage payment, then, you know, of course your wage would fall by $10 because you're helping the employee by hurting the employer, and you don't need a rocket scientist to tell you that that sort of thing would go on.

The idea behind equalizing wage differentials is if you have something that helps the employee, even if it doesn't hurt the employer at all, like sunsets become more beautiful in California, therefore, wages will fall in California because it just makes the job more attractive.  So in that sense, if we were in a unitary market, that critique would be killer because if you allowed the employee to recover, my scheme is, aha, the employee recovers but you're not hurting the employer because you're allowing the employee to bypass the employer and go to the distant manufacturer.  If the distant manufacturer could just pass it on, then that would be out of the question.

So what is useful or what I'm kind of assuming is that let's say Xerox Corporation is national and can't tell what state its customers are in.  So if New Jersey adopts a particular rule of tort law, then Xerox can't have discriminatory prices for clients that are in New Jersey versus clients that are in Texas.  For example, either you might say they don't know where their clients are, or maybe they do know where their clients are, but it would just be too easy for a New Jersey person to go to Texas, buy Xerox copiers at a low price and redistribute them in New Jersey or something like that.  So that may be an invalid assumption, but I am assuming that manufacturers have an imperfect ability to pass on those price increases to their employer clients.

MR.          :  Right.  And I think -- [Technical interruption.]  Got to record this for posterity.  Anyway, I'm not really sure I understand the price discrimination point because there's a higher cost for selling it to these guys when they can sue, and so I may sell it to some third party who then goes and sells it in the state, but they're not going to want to undercut me because presumably if they sell it, they're going to be liable.  And so it's just that the cost builds into this product whenever you sell in the state, and so I'm not sure why--personal price discrimination is not the right the right term because you're somehow implying different prices given the same cost, and it's not the same.

MR. VOLOKH:  I only meant having different prices.  Suppose that Xerox said, "Oh, AEI, you're in D.C.  D.C. has that tort law.  You have to pay more for your copiers to take into account tort law."  And AEI might go to Virginia and buy the Xerox machine at the low price and bring it back in.  So because of that possibility, I'm assuming that Xerox can't fully pass on the cost to you, the cost increase.  But maybe it would pass on--suppose we constrain Xerox to have a uniform price nationwide.  Then it might pass on a certain percentage of a cost increase to all of its clients nationwide, but basically the employer would be hurt somewhat by this, but not to the full extent.

So, you know, it's a simplifying assumption that I make, but it does seem realistic to me that while the product manufacturer can pass on some of the cost increase, it can't pass on all of it.

QUESTION:  Hi.  Eric Hellens [ph], Council of Economic Advisers.

Two comments.  One, at some point I actually think you're actually being too conservative in your standard errors.  I mean in effect you've moved something close to about 11 data points, which is you've gone down from one of the larger data sets I've ever seen to one of the smaller data sets I've ever seen.  So I think there are some ways you could sort of go around that, some John suggested, some others.

The second is, a lot of these states had tort reform at some point down the road, and so the second question was it might be something to look at going the other way and see if they got the wage premium back since there was a reverse of some of these particularly in the '80s.

Then the last point is with your industry dummy that you received so much grief about, so I apologize, but you have a problem here of crushing industry, that if these are the industries that made the product, get hit with a lawsuit, doesn't take too much of a deviation from sort of perfectly competitive labor markets to think that employees in those industries might bear some of the cost of their decline.  The one thing I would be worried about that industry dummy is that you're actually picking up the destruction of the general aviation industry and not necessarily a risk premium for people who fly airplanes in the course of their work now having free insurance.

MR. VOLOKH:  On the clustering, I guess I did--like I said before I did consider clustering on something like state here.  I suppose I was moved maybe--maybe I was inappropriately moved by this feature of clustering, that it doesn't actually change the value of--it doesn't actually change the coefficient, it only changes the standard error, so it's not really about if I had 11 data points, and still, 11 data points is under the extreme assumption that there was perfect correlation.  So I think the coefficients are whatever they would be.  The question is just it's kind of like a question of what p-value we're choosing, and so I suppose another way of taking this is that competitive negligence is the strongest effect that I found, and I have an argument about why it incorporates with the attenuation bias and how it incorporates all sorts of other omitted things.  Maybe I might defend that coefficient over the others on these other grounds.

You know, maybe you're right, that I ought to cluster by something bigger, and by something which would give me more clusterings or whatever you call that.

What was your second?

QUESTION:  What happens when the industry gets hammered [inaudible]?

MR. VOLOKH:  But that is what I'm trying to pick up.  By having a manufacturing dummy, I'm trying to take into account--the destruction of the general aviation industry is what I call a demand side effect.

QUESTION:  When you think about a demand side effect, what you're really not talking about is the impact of workers wanting to hire--I understood that to be I want to hire more people because my little commuter airline now doesn't have to pay the wage differential.  What you're really talking about is the fact that the industry is now basically facing an exogenous shock to its cost and may want to hire--

MR. VOLOKH:  Yeah, exactly.  The demand side effect is that the industry has lower labor demand because it's now facing these high costs, so that's the one thing that would lower wages.  And the supply side effect is the happy workers who say, "Nice lottery."  And so in any case, I'm with you on that one.

QUESTION:  My name is Mike--I'm unaffiliated, and if I could do away with tort law, as a person I probably would.  But my question comes to when you use the word "cost" and that's all you use and everybody uses the word "cost" let's just take the--

MR. VOLOKH:  Because I'm an economist.

QUESTION:  Let's just take the example of excess medical tests.  Then the x-ray people sell more x-ray machines.  There's more x-ray film sold.  There are more technicians reading it.  There is income offsetting or--not the word "offsetting" but the word "cost," means a sinkhole, but somebody's collecting those costs.  Is it a question of--

MR. VOLOKH:  Are you asking me or him?

QUESTION:  Well, Mr. Grace is the one that used the example.  What is the benefit or detriment of the system I think is the real point of it because I--you don't have to be a [inaudible] rocket scientist, but you don't have to be a rocket scientist to understand, yes, some workers benefit and some workers don't benefit.  The question is, what does it do to the system?

MR. GRACE:  Well, generally, when economists talk about costs they're talking about opportunity costs, and what we want to do is have things that have their highest value go to that value, and if we spend extra money on things we don't need we're wasting those resources, and that's a cost to society.  I mean that's the principle--

MR.          :  [Inaudible].

MR. VOLOKH:  I think what he's asking is that these guys, if left to their own devices, wouldn't have spent this money.  But they may have been willing to spend like 80 cents on the dollar or something like that.  So it's not--when you go and measure the full insurance cost, it's not all a loss in a sense.  You know, it may be like 20 cents on the dollar or 50 cents or 80 cents on the dollar, but it's something--

MR. GRACE:  But that 20 cents could have been spent on something the corporation wanted to spend it on instead of having to spend it on extra health premiums.  It might have been more valuable spent by the corporation or whoever on this, the society, than having spent on defense of medicine.  So it's a question of where is the value, and for a doctor who is risk averse from a lawsuit will spend a lot of extra money trying to avoid the lawsuit when he shouldn't have to.  I mean that's when I'm talking about--those kinds of costs, that's a loss in social welfare because it's spent on something that isn't really needed.

MR. VOLOKH:  Yeah, but it's not a loss per dollar spent because I still value the additional work that they're doing.  I just didn't value it at the full dollar that was there.

You could say that this one over-estimates.  As you say, there are lots of things that you weren't including in the cost too that are reasonable things to include, and you're underestimating for that too.

MR. KLICK:  Are there any other questions?

QUESTION:  Hi.  Real quick.  I'm Mary Beth from NFIB.  I was just wondering what the size of the corporations that you looked at were on average.

MR. VOLOKH:  I didn't look at corporations.  I looked at people who are in the current population survey.

MR. KLICK:  That's it.  We would like to thank all of our panelists.

In terms of a take-away thought that we can take from this panel, well, in general, as was my reason for inviting Marty to talk, too often our analyses of tort reform, either ex post or our thoughts about it going forward tend to be too simplistic, and we really do need to start thinking about these things more as a system, more in the general equilibrium approach, and I think Sasha's results speak to that effect.  And though Sasha was unwilling to draw any normative conclusions, I'm willing to draw one strong positive conclusion, is that there's no free lunch, and there are lots and lots of normative propositions that flow from that.

Thank you.

[End of program.]