Friday, September 18, 2009

Blast from the Past

One of the fears about lowering the drinking age is that traffic fatalities will rise. Usually, the pro-21 crowd either points to studies of the 1980s, when drinking ages were raised, or the 1970s, when drinking ages were lowered. Having already debunked the studies of the post-1976 period, using studies such as Miron and Tetelbaum (2009), here we look at a period with relatively scant data: 1970-1975, when 30 states lowered the drinking age.

Claim: There was an immediate and persistent increase in (presumably alcohol-related) fatalities among 18-20 year olds when the drinking ages were lowered (generally to 18).

First of all, we don't really know if that statement is even true to begin with, and we probably never will. Only a few decent-quality studies examined the years 1970-1975, the years in which the drinking ages were lowered. The Fatality Analysis Reporting System (FARS), which gives detailed reports about traffic fatalities, was not even created until 1975, and state-level data were not available through that system until 1976. So any conclusions drawn from state-level data for 18-20 year olds before 1976 is questionable at best.  For studies of nonfatal crashes, this caveat applies a fortiori. And any "alcohol-related fatality" data before 1982 is unreliable since FARS did not make this distinction until that year, which is understandable since a state that tested even 50% of fatal crash drivers for alcohol was considered stellar back then. Garbage in, garbage out.

From 1970-1975, any alleged increase in fatalities was imperceptible in the aggregate data. Using data from the National Safety Council, Miron and Tetelbaum (2009) showed that national 15-24 year old fatalities peaked in 1969, then declined sharply until 1975. From 1976-1980, fatalities rose somwhat, and declined from then on. The same was true for 18-20 year olds after 1976, when that group was separated out by FARS and the two groups have been highly correlated since. But the increase in the late 70s also occurred in states like California, which kept their drinking age at 21 throughout, so the 1976-1980 increase was unlikely to be a result of lowering the drinking age a few years prior.

So all state-level data for 18-20 year olds before 1976 must be gleaned from sources other than FARS, and some states had data problems for this period. And here's the grain of truth of it all. It is true that some states that lowered their drinking ages (and some that did not) saw increases in reported 18-20 year old fatalities from 1970-1975. But other states that lowered their drinking ages saw either no significant change or sharp decreases in such deaths in the table below:

StateDrinking Age Change (1970-1975)% Change in 18-20 total auto fatalities per capita (1970-1975)
Lowered, 21 to 19, 1975
Alaska*Lowered, 21 to 19, 1970no data
Arizona*Lowered, 21 to 19, 1972-29%
Arkansas21 (no change)-22%
California21 (no change)-14%
Colorado18 (no change)-10%
Connecticut*Lowered, 21 to 18, 1972+11%
Delaware*Lowered, 21 to 20, 1972+2.3%
DC18 (no change)no data
Florida*Lowered, 21 to 18, 1973-28%
Georgia*Lowered, 21 to 18, 1972-26%
Hawaii*Lowered, 20 to 18, 1972no data
Idaho*Lowered, 20 to 19, 1972-29%
Illinois*Lowered, 21 to 19, 1973-18%
Indiana21 (no change)-19%
Iowa*Lowered, 21 to 19, 1972, then 18, 1973-31%
Kansas18 (no change)-49%
Kentucky21 (no change)-31%
Louisiana18 (no change)-26%
Maine*Lowered, 20 to 18, 1972-14%
Maryland*Lowered, 21 to 18, 1974-8.6%
Massachusetts*Lowered, 21 to 18, 19730%
Michigan*Lowered, 21 to 18, 1972-6.4%
Minnesota*Lowered, 21 to 18, 19730%
Missouri21 (no change)-20%
Mississippi18 (no change)-46%
Montana*Lowered, 21 to 19, 1971, then 18, 1972+19%
North Carolina18 (no change)-10%
North Dakota21 (no change)+5.2%
Nebraska*Lowered, 20 to 19, 1972+7.8%
Nevada21 (no change)-61%
New Hampshire*Lowered, 21 to 18, 1973-59%
New Jersey*Lowered, 21 to 18, 1973+2.9%
New Mexico21 (no change)-14%
New York18 (no change)-9.7%
Ohio18 (no change)-35%
Oklahoma21 (not lowered to 18 until 1976)-14%
Oregon21 (no change)-13%
Pennsylvania21 (no change)-7.1%
Rhode Island*Lowered, 21 to 18, 1972+67%
South Carolina18 (no change)-25%
South Dakota*Lowered, 19 to 18, 1972-31%
Tennessee*Lowered, 21 to 18, 1971-1.6%
Texas*Lowered, 21 to 18, 1973+2.0%
Utah21 (no change)-49%
Vermont*Lowered, 21 to 18, 1971+161%
Virginia*Lowered, 21 to 18, 1974-17%
Washington21 (no change)-7.1%
Wisconsin**18 (no change for on-premise beer)**-7.1%
West Virginia18 (no change for beer)+1.9%
Wyoming*Lowered, 21 to 19, 1973+1.0%
(Taken from Cook and Tauchen (1984), Appendix A. Calculations ours. All data involves purchase age for beer unless otherwise noted. Dates taken from Wikipedia)

Thus, the state-level data are completely patternless, at least in terms of drinking age. Clearly, other factors were involved, such as gas prices (now known to have an effect), the economy, or even the weather. Only a handful of states (mostly with relatively small populations and hence much volatility in the numbers) in the table show significant increases, including one (North Dakota) that kept a constant 21 MLDA since the 1930s. The rest either saw sharp decreases or no significant change. This was in spite of the fact that, nationwide, the average driver in 1975 traveled more vehicle miles than in 1970. And before the advent of FARS, any increases are not clear as to whether they reflect true fatality increases or simply changes in how fatal crashes were reported. Nor does this table tell us whether those increases were contemporaneous with the age-lowering (done mostly in 1972-1973) since only two years, 1970 and 1975, were compared due to data availability. For example, Vermont's rather large increase, apparently, was not contemporaneous (See Douglass and Filkins, 1974).

A quick, albeit imperfect, way to estimate the effect of a policy change ceteris paribus is a method called "difference-in-differences," or DD for short. Here, we do a DD analysis comparing change vs. no change states:

Median difference, "change" states (1975 vs. 1970): -6.8%
Median difference, "no-change" states: -16.5%
Net difference-in-differences: +9.75%

This implies that, while both groups declined overall, the no-change states declined at a significantly faster rate than the states that lowered their drinking ages. And the latter group can be said to have more deaths at first glance. However, there are two radical outliers (VT and RI) that dramatically skew the results. Furthermore, Delaware is the only state that lowered the age from 21 to 20 and no further, and Wisconsin would better be included with the no-change states since 18 year olds were allowed to drink beer in bars both before and after. Thus, we omit the two outliers and Delaware entirely, and instead place WI with the no-change in our adjusted DD analysis:

Adjusted median difference, "change" states: -11.3%
Adjusted median difference, "no-change" states: -14.0%
Net difference-in-differences: +2.7%

Wow, that really makes a difference in the results. The net DD drops from nearly +10% to less than +3%. The latter "effect size" is small enough to be due to chance alone. Indeed, we also observe in the table that several of the states with significant increases are also states with some of the smallest 18-20 year old populations. Such states are prone to spurious shocks due to the volatility of smaller numbers of fatalities. Thus, we see that when we eliminate all the states with populations less than or equal to that of Montana, along with making the aforementioned adjustments, the drinking age effect disappears entirely:

Adjusted median difference, "change" states: -14.0%
Adjusted median difference, "no change" states: -14.0%
Net difference-in-differences: 0%

Regardless of what happened (or didn't happen) in the 1970s, it is essentially irrelevant today. Back then, drinking ages were lowered against a backdrop of falling real alcohol prices, higher adult per capita alcohol consumption than today, permissive and toothless DUI laws, social acceptability of drunk driving, no seat belt laws, ignorance about the risks of alcohol, and a generally cavalier attitude toward safety. The term "air bag" meant a person who talked too much. The term "designated driver" was not even in our vocabulary until the 1980s. Drunk driving was not just tolerated back then, it was expected of you if you were the least drunk person in the group (to drive everyone else home). Needless to say, things are very different today. So it's comparing apples and oranges. And any fear relating to the 1970s is therefore academic. Consider it debunked.

Claim:  There was a major increase in high school drinking (and related problems) when the drinking age was lowered, as 18 year olds bought for their younger friends.

Again, correlation does not equal causation.  It was true that in the 1970s, teen drinking increased, but that was a national trend that occurred in essentially every state, including those like California that kept the drinking age at 21.  In fact, the secular trend predated the 1970s by many decades, and ironically enough began during Prohibition in the 1920s.  It lasted until about 1979, then the trend reversed and teen drinking declined through the 1980s until the early 1990s.  And the downward trend predated the raising of the drinking age, and again occurred in essentially every state.

The average age at first drink did decline nationwide, but that began in 1965 (or earlier), and continued to decline long after the drinking age was raised in the 1980s. And California, who had a 21 drinking age since 1933, saw the same trend overall as the rest of the country. So clearly other factors are at work, and the trends cannot be traced to changes in the drinking age.

Part of the increase in teen drinking could be that parents stuck their heads in the sand about alcohol, being relieved that "at least my kid isn't smoking pot."  Or it could have resulted from a moral panic and a consequent deviancy amplification spiral.  Or perhaps a bit of both, with the former preceding the latter.

Interestingly, a 1977 study found that high school seniors in states with a drinking age of 18 actually drank less and had fewer alcohol-related problems than those in states with a drinking age of 20 or 21.  Why this is is not entirely clear, but the researchers hypothesize that "forbidden fruit" may very well entice those in the more restrictive states to drink.  Or perhaps those seniors that are still 17 are more likely to wait until 18 to be legal since this is more realistic than waiting until 20 or 21.

Again, much of the fears from the 1970s are now academic, and are unlikely to be a problem if the drinking age was lowered today.  The notion of teenagers having "liquid lunches" in high school is no longer socially acceptable--it is now considered a sign of a drinking problem.  Those who are caught bringing booze to school are dealt with much more harshly than they were back then, campuses are often closed, and students are essentially defanged and declawed.  And today's tough enforcement requires IDs to be shown when purchasing alcohol, reducing the chances of a 15-17 year old "passing" for 18.  About the last remaining fear is high school keggers (which still occur even with a 21 drinking age), but any possible increase in these parties can (by definition) be prevented by keeping the purchase age at 21 (or 20) for bulk quantities like kegs and cases while lowering it to 18 for everything else. 

One thing, however, is for sure:  teenagers will get their hands on booze one way or another, and whether the drinking age is 18 or 21 is of little consequence to this fact.  Where there's a will, there's a way.  And where there's a swill, there's a sway.  So consider this one debunked as well.


No comments:

Post a Comment