safety[f82,jmc] When safety kills

        It is a truism that poverty kills, and we can try to quantify
this proposition statistically by computing the relation  between the
income of  a state  and its  death rate  - both  taken from  the 1979
Statistical   Abstract   of  the   United   States.    This  probably
uncontroversial proposition has a probably controversial  corollary -
too much money  spent on a safety  measure kills more than  it saves.
Namely it drains money out  of the economy that the people  who would
otherwise get the  money and be more  prosperous would spend  to live
longer.

        The estimated cost  of safety measures  in lives is  one life
for several million dollars spent.  The number, as we  shall explain,
cannot be computed accurately, but all our estimates are  between one
million and five million dollars.  This means that if more  than five
million dollars  are spent on  a safety measure  per life  saved, net
lives are lost.  This relates  to social policy, because many  of the
required  safety  measures   for  nuclear  power  plants   and  other
industrial facilities may cost many tens of million dollars  per life
saved.  Since the money spent  on safety comes out of  utility rates,
it seems to follow that the public would be safer if  the expenditure
were not made and rates kept down instead.

        The  same  appears to  be  true of  the  $100  million recall
campaign for Tylenol unless it is believed to have saved more than 20
lives.   It may  even  be true  of the  tamper-proof  packaging being
introduced.

        The statistical technique is quite simple and even naive, and
the  numbers  obtained   are  certainly  subject   to  qualification.
However, some of the safety expenditures are so far out of  line that
the conclusion that they  are irrational isn't very sensitive  to the
numbers.

        Our data comes from  the "Statistical Abstract of  the United
States:  1979" (100th  edition), U.S.  Bureau of  Census, Washington,
D.C., 1979.  Table No. 107 on  page 74 gives data on death  rates for
each state  and also grouped  into nine regions  of the  country.  We
used  the last  column which  gives the  death rate  per  thousand of
population for  the year  1977, the latest  included.  For  income we
used Table No. 730 on page  445 which gives the per capita  income by
states and with  the states grouped into  nine regions.  We  used the
column for 1978, the most recent given.

        For doing the computation, we used the  Hewlett-Packard HP-15
hand-held calculator and its built-in program for  linear regression.
We did the computation by regions first and obtained the formula



        <death  rate  (per  thousand)> =  10.46  -  0.00023 <personal
income>.

        Let's abbreviate this

        d.r = 10.46 - 0.00023 p.i.

        Remembering that  the death rate  is given per  thousand, the
conclusion is  that for  each $4.4  million of  personal income  in a
state in 1978, one fewer person died in 1977.  We're sorry  about the
fact that the years don't  agree exactly, but that's what was  in the
tables.  The reader with access to other data can probably  find data
for matching years.

        The correlation co-efficient  was -0.2, which  indicates that
income accounted for 20 percent of the variation of death  rate among
regions.

        We also  did the calculation  in three other  different ways.
Taking the  five highest and  the five lowest  states in  income, the
calculation gave

        d.r = 13.6 - 0.00064 p.i

        with a  correlation of  -0.63.  An  income decrease  of $1.56
million seems to cost a life.

        The  highest  income  states  included  Washington  D.C.  and
Alaska.   Washington  has  the  second  highest  death  rate (perhaps
because it has  a high income segment  of Government employees  and a
low income group of poor  blacks), and Alaska has the  highest income
and the lowest death rate (presumably because its elderly  retire and
die  elsewhere).   If  we eliminate  D.C.  and  Alaska,  the equation
becomes

        d.r = 13.3 - .0006 pi,

with a correlation of  -0.82, which isn't much different.   An income
decrease of $1.6 million seems to cost a life.

        Taking the five highest and five lowest death rates gies

        d.r. = 14.5 - .000846 p.i.

with a  correlation of  -0.29.  An income  decrease of  $1.18 million
seems to cost a life.



        Finally, we did the calculation with all 50 states  plus D.C.
which gives

        d.r. = 11.8 - .00043 p.i.

with a correlation of -0.35.  This suggests that fully a third of the
variance in death rate among states is accounted for by variations in
personal income.  Dividing the 0.00043 by 1000 and inverting  gives a
result suggesting  that every  reduction in  personal income  by $2.3
million costs a life.

        Discussion:

        If one  wanted an accurate  answer, a  prospective experiment
would be needed,  i.e. one should  have an experimental  group (given
money) and  a control  group left  alone.  Ideally,  the experimental
group shouldn't  know they  were being  subsidized -  their employers
might be subsidized to increase their pay and give misleading reasons
for doing so.  Such an experiment is unlikely to be performed,  so we
are reduced to inferring what we can from the statistics available.

        The interesting number from  the policy point of view  is the
amount of decrease in  personal income that apparently costs  a life.
The estimate ranges  from the $1.18 million  when we take  the states
with highest and lowest death rates to the $4.4 million obtained when
we  aggregate by  regions.  Asked  to guess,  I'd go  for  the number
obtained when all states are taken into account, namely $2.3 million.

        I have  consulted professional  statisticians, who  point out
various possibilities for quantitative error.  Some errors  may arise
from  the  aggregation  by  states,  and  other  may   give  spurious
correlations,  e.g.  a  state  to which  many  people  retire  may be
expected to have  a lower income and  a higher death  rate.  However,
no-one has suggested that another technique or more data is likely to
give qualitatively different results.

        We may  speculate about how  income saves lives.   People eat
better, value their  lives more and  avoid some risks,  visit doctors
more  readily,  and  pay  more taxes  for  public  health  and safety
measures.   All  these factors  combine  in producing  the  effect of
average personal income  on average death  rate.  For the  purpose of
the  present  analysis,  it  isn't  necessary  to  distinguish  these
factors.  This is fortunate, since the information required to  do so
probably isn't available.

        There are other ways of comparing money and lives.   First of
all, we may look for ways of saving lives with money.  Many years ago



it was  estimated that $100,000  spent judiciously on  stoplights and
other  automobile safety  measures  would save  a life  year.   I was
recently informed that the most cost effective way of saving lives in
the U.S. is to increase paramedic services.    Most likely  lives can
be   saved  even   more  cheaply   by  public   health   measures  in
underdeveloped  countries.  However,  the advocates  of  an expensive
safety measure  can point out  (if they are  inclined to  address the
issue at all)  that there is no  guarantee that money unspent  on one
safety measure will be spent on another more cost-effective measure.

        Yet  another  way  is  to  estimate  the  risks  people  take
voluntarily to  get money  or to  save it.   These also  give smaller
estimates of the monetary value of life than the cost of  many safety
measures.   The advocates  of a  particular expensive  safety measure
often point  to the  difference between  risks taken  voluntarily and
those imposed by others.  Starr (19xx) discusses these issues.

        All  these  approaches  are relevant  to  deciding  whether a
proposed expenditure  is worthwhile.   However, the  present approach
has  the  advantage  that  it  compares  an  action  (undertaking  or
requiring an expensive safety measure) with doing nothing - an option
that is often available to a decision maker, who usually doesn't have
the option  of an  alternate expenditure on  saving lives.   Thus the
Nuclear Regulatory Commission cannot  require that a utility  build a
clinic  in  Indonesia  instead  of  making  the  earthquake  proofing
stronger,  but often  it  has the  the  option of  not  requiring the
expenditure.  So, often, does a judge.

        Considerations of this kind are most important if they affect
policy, and  affecting policy means  politics, and  politics involves
expressing conclusions in  politically effective ways.  We  offer two
ways of putting the conclusions of this study - mild and sharp.

        The mild way of putting it is that policy makers who consider
a  particular  expenditure  on  safety  -  whether  it  is  their own
organization's  money  or  whether  are  in  a  postion   to  require
expenditure by others - consider whether the expenditure will  save a
life for every  $2.3 million spent.  If  not the drain they  cause on
the economy may cost more lives than the safety measure saves.  Meta-
policy makers, i.e.  Congress and state legislatures,  might consider
requiring impact statements estimating  lives to be saved  per dollar
spent on specific safety measures before a regulatory body can impose
a new safety measure.  It might be remembered that it isn't  easy for
a regulator faced with TV broadcasts, press conferences by activists,
and questions by excited legislators to say that the  proposed "tough
new regulatons" won't save as  many lives as they will cost,  and the
best option  may be  to do  nothing.  It  is especially  difficult if



doing nothing puts the value of the regulator's own job  in question.
For this reason,  a meta-rule requiring  the impact statement  with a
$2.3 million cost limit per life saved may be helpful.

        The sharp way of putting the issue is to accuse  Ralph Nader,
the  Natural Resources  Defense Council,  the Sierra  Club, Musicians
United  for Safe  Energy,  the blockaders  of the  Seabrook  site and
Diablo Canyon, the California Legislature, the Washington, D.C. Court
of Appeals and  other busybodies of having  killed a large  number of
people in furtherance of arbitrary notions of who were the  good guys
and who  were the  bad guys.          Here  are some  of the  ways in
which the conclusions might be attacked:

        1. The most drastic  attack is to suggest that  the causality
is reversed.   Perhaps states  are healthy  or unhealthy  for reasons
unrelated to income, but  healthy states have higher  income, because
sick people don't earn as much money.  The only reason  to disbelieve
this is that plausible reasons why specific states are healthier than
others are hard  to think up, whereas  the economic reasons  why some
the presence of  specific economic opportunities make  certain states
wealthier readily come to mind.