Sunday, April 17, 2011

No Effect, No Cause III — The Sequel Returns

Over the last couple years (here and here), I have banged on about studies purporting to show some greatly elevated crash risk from cell phone usage despite seemingly contradictory reality:
At the time, my main beef was that The Studies [asserting a 400% increase in crash risk due to cell phone usage] had assumed their conclusion before so much as getting their analytical key into the ignition. The fact of a declining accident rate in the face of skyrocketing cell phone usage would seem worth something more than deep silence: by assessing risk without regard to actual consequences, they were trafficking in crimes against statistics.

Secondarily, and to some very well mannered derision, I surmised that cell phone usage simply does not cause enough accidents to be statistically noticeable.

As it happens, my original surmise might have been close to the mark was right on the money.
A recent study by Vikram Pathania of the London School of Economics and Saurabh Vikram of the University of Chicago substantiates my original assertion: If risk does not reflect in rate, then risk becomes a cause without an effect.
The link between cell phone use while driving and crash risk has in recent years become an area of active research. The most notable of the over 125 studies has concluded that cell phones produce a four-fold increase in relative crash risk comparable to that produced by illicit levels of alcohol. In response, policy makers in fourteen states have either partially or fully restricted driver cell phone use. We investigate the causal link between cellular usage and crash rates by exploiting a natural experiment induced by a popular feature of cell phone plans in recent years'the discontinuity in marginal pricing at 9 pm on weekdays when plans transition from 'peak' to 'off-peak' pricing. We first document a jump in call volume of about 20-30% at 'peak' to 'off-peak' switching times for two large samples of callers from 2000-2001 and 2005. Using a double difference estimator which uses the era prior to price switching as a control (as well as weekends as a second control), we find no evidence for a rise in crashes after 9 pm on weekdays from 2002-2005. The 95% CI of the estimates rules out any increase in all crashes larger than .9% and any increase larger than 2.4% for fatal crashes. These estimates are at odds with the crash risks implied by the existing research. We confirm our results with three additional empirical approaches: we compare trends in cell phone ownership and crashes across areas of contiguous economic activity over time, investigate whether differences in urban versus rural crash rates mirror identified gaps in urban-rural cellular ownership, and finally estimate the impact of legislation banning driver cell phone use on crash rates. None of the additional analyses produces evidence for a positive link between cellular use and vehicle crashes.
Emphasis added.

There are at least several lessons here. First, for true believers, no amount of contradictory evidence will budge their belief. Which, in turn, ensures that pointless laws criminalizing behavior without notable harm will endlessly fester on the books. Most importantly, though, is the corollary with warmenism, which really amounts to a cause unashamedly adamant, despite being suspiciously disconnected from meaningful effects.

14 Comments:

Blogger Bret said...

Absence of evidence is different than evidence of absence.

You're reading a bit too much into the study, IMO.

April 17, 2011 3:35 PM  
Blogger erp said...

Around here I rarely see people driving with a cell phone to their ears anymore. Apparently, ear pieces and phones built into cars have become more popular now.

April 17, 2011 3:37 PM  
Blogger Mark Frank said...

This is just one of many studies into the link between mobile phone usage and accidents. The results vary from this one (no evidence of a significant difference) through to others which show an increased risk while making a call by 4 or 5 times as much. (It is easy to confuse this with an increased risk while driving of 4 or 5 times - but it is not all the same thing.)

Why do you believe this study trumps all the others instead of looking at a cross-section of studies? The authors write: "Given the strength of our research design, we believe that our paper may meaningfully add to the discourse regarding the efficacy of policies restricting driver cell phone usage." i.e. they accept they are not the only game in town.

April 18, 2011 9:39 AM  
Blogger Hey Skipper said...

Bret:

Yes, absence of evidence and evidence of absence are two different things.

In this case there is substantial evidence of absence: actual accident trends from which one would never guess cell phones were ever invented; and now this ingenious study which leaves the strong impression that accident rates among cell phone users are indistinguishable from the rest of us.

---

erp:

You are right. On both my older cars I have installed sound systems (at a mere $200 a whack) that include Bluetooth hands free calling.

I still see people using cellphones -- this is Alaska, after all -- but more and more I see people about whose sanity we would be wondering in an earlier era.

---

Mark:

Why do you believe this study trumps all the others instead of looking at a cross-section of studies?

Primarily, because this study is consistent with reality. Cell phones have gone from non-existent to pervasive in twenty years. Yet over that same period, accident rates have declined despite their putative operational risk.

Second, it wasn't some laboratory study, simulation, or setup where the researchers observed and categorized driver behavior (what could possibly go wrong there?) for a small number of behaviors. Instead, with a very large N (440,000) it compared the phenomena in question with a control group. That is kind of the gold standard for experiment design, isn't it?

As for distinguishing cell dialing from conversation, the amount of time spent doing the former is so small that one could toss out any risk value shy of total without having to worry about further explanation.

Maybe it is just me, but if I had a theory whose consequences run counter to reality, I'd think that rather worth some explaining. So far as I know, studies proclaiming risk similar to driving under the influence have not troubled themselves with this bothersome inconsistency.

What I think the "researchers" are missing, and which slobbering legislators won't possibly ascertain, is the notion of compensatory behavior.

Snowy roads are far riskier than dry. Drivers compensate for that risk by slowing down, increasing following distance, minimizing lane changes, etc to the degree required to make the perceived risk equal to that of a "normal" day.

Similarly with cell phone usage -- people adjust their driving to compensate so that their net level of risk remains the same.

My larger point goes to warmenism. For any hypothesis wishing to become a theory, it must have at least one deductive consequence.

Studies "demonstrating" some elevated risk from cell phone usage have a single deductive consequence: accidents.

Which appear absent; hence, the theory appears negated.

What are AGW's deductive consequences?

April 18, 2011 11:32 AM  
Blogger Bret said...

Hey Skipper wrote: "actual accident trends from which one would never guess cell phones were ever invented"

It's not like one day nobody had a cell phone and the next everybody did. It was a gradual trend which I doubt could be detected in the accident rate trend whether or not cell phone usage has an impact.

Also, this "ingenious study" at best shows lack of evidence that driving while talking on a cell phone during off peak hours is a problem. At 9 PM the traffic volume is pretty low so I can believe cell phone usage wouldn't have much impact then.

April 18, 2011 11:59 AM  
Blogger Mark Frank said...

Hey Skipper

Oh dear - we have been over this before. Studies which crudely correlate accident rates against cell phone usage are useful but have problems. They are confused by other confounding effects and even a big increase in risk from mobile phones may not show up.

The authors say:

"Given the RT estimates of relative crash risk, the size of the observed discontinuity in call volume, and a conservative assumption of driver cell phone use, we would expect to see an 2 to 6% rise in crashes in the hour following the threshold."

i.e. even if using a cell phone increases the risk by 4 times (the RT estimate) then the effect on accidents would only be 2-6%. Note that the authors results were compatible with a 2.4% increase in fatal accidents (95% confidence interval).

More focussed studies looking at such things as accident reports avoid the problem of confounding effects but also have their problems - for example they omit compensatory behaviour (do you really think professional researchers have not thought of this?).

There is scope for both types of study.

AGW's deductive consequences are rising global temperatures - which, contrary to rumour continue to rise. But this has been debated about a million times on the internet and I don't intend to repeat it.

April 18, 2011 12:13 PM  
Blogger Hey Skipper said...

Mark:

Note that the authors results were compatible with a 2.4% increase>

Correction — at the 95% CI, the results rule out any increase in fatal crashes greater than 2.4%. While I don't know the confidence interval, I think it is very likely the results are compatible with an increase of fatal crashes of anywhere between 0% and 2.4%.

Particularly given that the increase in all crashes was, at most, 0.9%. If that is true, then cell phones would have to be wildly over represented in fatal mishaps. Possible? Yes. But I'll bet it has a heck of a lot more to do with the size of the confidence interval.

The real, and irrefutable problem, is that the hypothesis that cell phone usage is a cause with an increase in mishaps as a consequence explains observations no better than the null hypothesis (i.e., there is no relationship between cell phone usage and mishap risk).

I think that researchers who trumpet their findings of greatly increased risk ignore or don't think about compensatory behavior. If they did, they would need to do something else with their time.

AGW's deductive consequences are rising global temperatures - which, contrary to rumour continue to rise.

AGW's hypothesis is that rising CO2 will lead to rising temperatures. An example of a deductive consequence would be, for instance, more and stronger hurricanes.

I don't doubt that increased CO2 leads to an atmosphere that is warmer than it would be otherwise. However, if the consequences stop at that first order effect, then AGW is a total non-event.

It is the second order consequences — more and stronger hurricanes — that make AGW a big deal.

Which is the problem: granting CO2 causes warming (cell phones increase the risk of a crash) what should happen as a result (where are all the crashes)? If climate has compensatory behavior, then why give a darn about AGW?

(BTW, the rate of warming since 1998 is essentially zero.)

April 18, 2011 3:57 PM  
Blogger Mark Frank said...

This comment has been removed by the author.

April 18, 2011 10:35 PM  
Blogger Mark Frank said...

Hey Skipper:

While I don't know the confidence interval, I think it is very likely the results are compatible with an increase of fatal crashes of anywhere between 0% and 2.4%.

Therefore the results are compatible at the 95% level with a 2.4% increase in fatal accidents.

The real, and irrefutable problem, is that the hypothesis that cell phone usage is a cause with an increase in mishaps as a consequence explains observations no better than the null hypothesis (i.e., there is no relationship between cell phone usage and mishap risk).


For this particular study I think this is true. But don’t oversimplify what it is saying. If the increase in risk were 4 times (and a number of other assumptions hold) then there is a 1 in 20 chance that we would get an increase in accidents as low as this or lower. If the risk is in fact lower then the probability of getting such a low result increases. Also bear in mind that even if the increase in risk were 4 times, then 1 in 20 studies would show figures as low as these. Ideally you would take a Bayesian approach and take into account the prior probability of the alternative hypotheses – which would be based on other studies. But I don’t see any attempt to do this.

I am not dismissing this study. I haven’t read it in detail but it appears to be a very professional piece of work. But it is only one study. To place so much emphasis on it betrays a political bias. You like the conclusions of this one so it trumps all others.

AGW's hypothesis is that rising CO2 will lead to rising temperatures. An example of a deductive consequence would be, for instance, more and stronger hurricanes.

I don't doubt that increased CO2 leads to an atmosphere that is warmer than it would be otherwise. However, if the consequences stop at that first order effect, then AGW is a total non-event.
It is the second order consequences — more and stronger hurricanes — that make AGW a big deal.


OK. Let’s go with rising sea levels. They follow almost certainly from rising temperatures. According to satellite measurements average sea levels have risen about 60 mm in the last 20 years. A rise of 400 mm in the Bay of Bengal would put 11 percent of the Bangladesh's coastal land underwater.

(BTW, the rate of warming since 1998 is essentially zero.)

It all depends on where you start and how much smoothing you do. Why not look at a well-known sceptic’s chart:

http://www.drroyspencer.com/latest-global-temperatures/

and see what you think . 10-15 year trends are really too short. See how much depends on single year - 2008

April 19, 2011 1:08 AM  
Blogger Hey Skipper said...

Bret:

It's not like one day nobody had a cell phone and the next everybody did.

Not one day, but it has been a huge leap in a decade: 34 million to 203 million between 1999 and 2009, which by now amounts to a seven fold increase.

At 9 PM the traffic volume is pretty low so I can believe cell phone usage wouldn't have much impact then.

It is also dark. From a Forbes article "Most Dangerous Times to Drive": Time of day plays an important role in evaluating fatal crashes, in no small part because other dangerous factors are compounded at night. The instances of drunk driving, speeding and driving without a safety belt all significantly increase during the night hours and each contributes directly to increased fatality rates.

Mark:

Therefore the results are compatible at the 95% level with a 2.4% increase in fatal accidents.

I think the better way to read that is that the results exclude anything greater than a 2.4% increase. 2.4% isn't what it is, just the greatest it could be.

Particularly when the preceding sentence says "we find no evidence for a rise in crashes after 9 pm on weekdays from 2002-2005." Also, the one following "These estimates are at odds with the crash risks implied by the existing research."

I am placing this much emphasis on this study for several reasons. It is the second one I know of with the same result (the prior one found absolutely no change in mishap rates in states where cell phone use was banned, despite apparent widespread compliance). Second, and most importantly, the study's findings are consistent with the reality of declining crash rates over the entire period cell phones have become ubiquitous.

The dog that hasn't barked needs explaining.

OK. Let’s go with rising sea levels. They follow almost certainly from rising temperatures. According to satellite measurements average sea levels have risen about 60 mm in the last 20 years.

Equally, we could go with retreating glaciers (which probably explains rising sea levels better than ocean temperature).

Unfortunately, both precede the 20th Century's rise in CO2.

It all depends on where you start and how much smoothing you do. Why not look at a well-known sceptic’s chart: and see what you think . 10-15 year trends are really too short. See how much depends on single year - 2008

You are right, 10-15 year trends are too short, and I don't mean to imply that we can take the last ten years as proof of anything. However, since 1998, the rate of warming, even dropping out 2008 entirely, is very low.

As with cell phone risk, the consequences (absent the first order effect) are very difficult to ascertain.

April 20, 2011 4:08 AM  
Blogger Mark Frank said...

Hey Skipper


me:

OK. Let’s go with rising sea levels. They follow almost certainly from rising temperatures. According to satellite measurements average sea levels have risen about 60 mm in the last 20 years.

you:

Equally, we could go with retreating glaciers (which probably explains rising sea levels better than ocean temperature).

Unfortunately, both precede the 20the Century's rise in CO2.


You already accepted that CO2 will lead to an atmosphere that is warmer that it would otherwise be.


Do you dispute the rising temperatures cause rising sea levels? This happens through thermal expansion and loss of ice from land (including glaciers).

If you accept this then logically you must accept that increasing CO2causes sea levels to be higher than the otherwise would be.


Average temperatures have been rising since the little ice age for reasons unconnected to CO2 increases and consequently there has been a rise in sea levels.

April 20, 2011 12:19 PM  
Blogger Hey Skipper said...

If you accept [that increased CO2 will lead to a warmer atmosphere] then logically you must accept that increasing CO2 causes sea levels to be higher than the otherwise would be.

I accept that, as a first order effect, increasing CO2 will lead to a warmer atmosphere, but the amount is so small as to be ignorable.

The problem with choosing either sea level or glaciers as signs of AGW is that they have been, respectively, rising and shrinking for far longer than anyone has been able to measure such things.

As it happens (and perhaps this is the sort of thing I shouldn't admit in public), I have been reading Gibbon's Decline and Fall of the Roman Empire. About midway through Volume I, he goes into some length about the historical record regarding climate change, with clear evidence that it had swung between cold and warm over a matter of several centuries.

Both these things present the same problem. SFAIK, climate science has no explanation for either the LIA or MWP, yet insists that it understands all the non-CO2 forcings, and hence, by elimination, the CO2 contribution to current climate change.

However, that amounts to saying that increased CO2 is significantly explanatory when the null hypothesis says otherwise.

Which points to my analogy to a putative quadrupling of risk due to cellphone usage. In a relatively simple system containing a variable that is supposedly significant, and rapidly changing in a short period, there should be an easily distinguishable consequence provided that simple system changes mechanistically.

Yet both reality — actual mishap rates — and large scale studies directly contradict that supposition. This, in turn, raises the question: what observation would be sufficient to conclude that an increase in cell phone usage will not lead to an increase in mishaps?

This is where I analogize to CAGW. SFAIK, there are no deductive consequences — other than post-hoc ones such as sea level rise or glacial retreats — attending CAGW that have actually panned out. Take, for example, the UNs assertion that by 2010 there would be 50 million climate refugees. (Recently, without anything like the original fanfare, changed to 2020.)

So, as with cellphones, what climate observation would suffice to disprove CAGW?

If the answer is the null set — as I suspect it is — then people are assuming the conclusion merely because the premise is intuitively obvious, instead of there actually being a cause-effect connection.

April 23, 2011 8:11 PM  
Blogger erp said...

Skipper, the same dire warnings were posited when the execrable 55 MPH laws were rescinded. The pesky facts remain that accidents didn't go down when Nixon imposed them and didn't go up when they were lifted -- causing a lot of consternation among nanny-staters.

April 24, 2011 6:28 AM  
Blogger Hey Skipper said...

erp: Thanks for reminding me -- I lived through that nonsense, too.

Repealing the NMSL, was supposed to cause our highways to run red with blood.

Instead, accident rates and fatalities have been on a downward trend ever since.

April 24, 2011 2:49 PM  

Post a Comment

<< Home