APROPOS OUR DISCUSSION LAST EVENING ON THE QUESTION OF THE

• Document Type: CREST
• Collection: General CIA Records
• Document Number (FOIA) /ESDN (CREST): CIA-RDP85T00153R000100030005-3
• Release Decision: RIFPUB
• Original Classification: K
• Document Page Count: 4
• Document Creation Date: December 20, 2016
• Document Release Date: January 14, 2008
• Sequence Number: 5
• Publication Date: February 25, 1983
• Content Type: MEMO

Approved For Release 2008/01/14: CIA-RDP85TO0153R000100030005-3
25 February 1983
NOTE FOR: DCI
DDCI
DDI
NIO/USSR-EE
NIO/SP
FROM : C/NIC
1. Apropos our discussion last evening on the question of the
intelligence community performance in relation to our "triad" of
strategic/arms control estimates, I believe that the principal failing
was in redictin the magnitude of the Soviet missile buildup during the
1960s an through 1972. In contrast the monitoring performance has been
good.
2. Albert Wohlstetter documented the failings of the predictions
in two-articles in Foreign Policy magazine in 1974 and 1975. Attached
are the relevant graphs of the forecasts from these two articles.
Henry S. Rowen
Att: a/s
Approved For Release 2008/01/14: CIA-RDP85TO0153R000100030005-3
 Approved For Release 2008/01/14: CIA-RDP85TO0153R000100030005-3
It is a good idea, then, to subject to sys-
tematic test this claim of regular overestima-
tion. Its nearly universal acceptance has
emerged from constant repetition of tags like
"we are racing ourselves," 4 rather than from
any numerical comparison of estimates with
reality. Figures 1-3 illustrate and Tables I
and 2 sum up the results of a search o he
Secretary of Defense's annual Posture State-
ments from 1962 tt 11Q72 for all Irmo-term
predictions of Soviet strategic missile and
bombes deployments.. and a comparison of
them with what the Russians actually de-
ployed by mid-1972, the last date referred
to in the predictions that could then be
checked. I use posture Statements rather
than Intelligence: first because the Secretary
made quite precise predictions. second be-
cause he assured us that the high end of his
range was higher than the highest in the na-
tional intelligence estimates: and third be-
cause he used his forecasts directly to support
defense programs, and so they are more rel-
The first three charts, Figures IA to IC.
compare some U.S. predictions of Soviet
ICBM launchers to be deployed with the ac-
tuality as estimated after'the fact. The verti-
cal arrows show when the :prediction was
made (e.g.. February: 1962 in Figure IA).
The dashed line or lines indicate the range
from high to low of what was predicted.
(In Figure IA. a high of 650 and a low of
350, by mid-1967, five-and-a-half years
,later.) Later forecasts *usually included (as in
..Figure 1B) a high and a low for more than
one year. This is shown in the shaded por-
tion. The steeply -rising solid line, which is
the same in all the charts, shows the number
the Russians actually completed, as ed
Though the'claim about invariable over-
estimation posits that at least the middle of
the range between high and low always ex-
e.q.. Lipton and Rodberg. op. cit., p. 103; Wiesner.
ABM: Yes or No, p. 18: Pano/sky, "Roots of the
Strategic Arms Race: Ambiguity and Ignorance." Bul-
letin of the Atomic Scientists. June 1971, p. 15.
!'redictio de short-term estimates that are lim-
ited euenticily to the completion of launchers already
started. "Actual number" refers to og%e al estimates
made after the ICBM's were deployed.
Approved For Release 2008/01/14: CIA-RDP85TO0153R000100030005-3
 Approved For Release 2008/01/14: CIA-RDP85T00153R000100030005-3
n.7 ~/ N N N U N N
>7177?
ISSEt exclude short-t erm estimates that are limit.
started. essentially to the completion of launchers already
.
reeds 'reality, it will be apparent that even
the high end of the range seldom did that.
and then only at the start of the period-
and even then just barely. For ICBM's, the
"highs" reached as high as realit on twice
in es. e
pre tction. made in 1965
is qui a typical. Figures 2 and 3 illustrate
analogously typical long-run predictions of
future Soviet submarine-launched missiles
deployed and future Soviet bomber deploy-
ments. The middle of the predicted range of
the number of sub-launched missiles de-
ployed was about three-fourths of the even.
Wohlstetter
tual reality. In the case of the bombers, we
continued to believe that the Russians were
going to phase them down and most dras-
tically in the case of the medium bombers;
but the Soviets never came down to our ex-
pectations. Tables I and 2 sum up some
principal results. Out of 51 predictions, the
low end of the range never excee e the ac-
ual. the mean between a tg an low
e ded it only twice in 51 times; put highs
reached reality only nine times! H-a y a
record o overestimation: orover, the ra-
tios of predicted-to=actual future values of
the Soviet strategic force in operation display
the fact that the underestimates were very
substantial,and that, even the average.of the
highs was under the reality Analysis also
makes it evident that there was no systematic
learning tom the 'formation ac-
fpmulated.
Infact, nce the estimates shown refer
to the cumulative number. of strategic vehi-
cles in operation at' future.,dates,.and since
later predations were based on more exten-
sive knowledge of what was already de-
ployed or at least started:`in. construction at
the time of. the prediction. the degree of bias
can be made even plainer:-';_'.
First. our means of. acquiring information
improved greatly over the_period. Second, in
later. years a much larger~proportion of the
cumulative total in operation was alread
i
y
n
operation at the time predictions were made.
And third, ~-e had information not only
?lwut the number of launchers completed
and in operation (displayed in the rising
curv
f
es o
Soviet ICBM and SLBM launchers)
but also about the substantial numbers of
a~ uiaue.
e knew that ICBM's started would
l
genera
-
ly be compIeted. sayJn about a yea _ n half. nd submarine-based missile launchers
to a ut two-and-a-half years. but in any
case well before the dates in our long-run
predictions. In fact, estimates of the missile
launchers already started that were expected
to be completed by a given time averaged,
Approved For Release 2008/01/14: CIA-RDP85T00153R000100030005-3
 Approved For Release 2008/01/14: CIA-RDP85TO0153R000100030005-3
Figure 3
Air Force and official consensus predictions
made in fall 1962
prediction
vyes made
Figure 4
Air Force and official consensus. predictions
nude in fall 1966
Date
prediction
was made
Mid-calendar year
Estimated actual
inventory
Air Force
prediction
Official consensus
prediction
Estimated actual
inventory
Air Force
prediction
Official consensus
prediction
60 62 64 66 68 70 72? 74 76 78
Mid-calendar year
Wohlstetter
below the 1967 reality and the "high"
barely reached it. In fall 1963, the Air Force
predictions still greatly exceeded the con-
sensus, but the two began to converge.
There was some overlap between them in
the early years referred to in the prediction,
and in the more distant years, where the Air
Force outbid the consensus, even its high
dropped below reality. In fall 1964 the Air
Force and official predictions came close to-
gether and overlapped for the first time in
predictions about the more distant years.
For these more distant years, even the Air
Force highs were below reality,.though the
Air Force still exceeded the consensus. In
fall 1965 and 1966 (Figure 4) underesti-
mation worsened -with "further . convergence.
Finally, in fall 1967, convergence was total.
The Air Force endorsed'
' .the consensus on
condition that the Soviets would deploy
MRVs (Multiple Re-entry Vehicles-unlike
MIRVs, not aimed independently), which
they did. The highs of the long-term. fore-
casts in these last years through 1967 were
invariably under reality,.and both the con-
sensus and the Air Force assumed an ulti-
mate leveling off of the Russian. program
well below what happened. In fall 1968
the Air Force concurred with the consensus
on the assumption, now clearly conservative,
that MIRVs would be deployed by mid-1978.
The steady movement toward the official
forecasts suggests the power of consensus.
That power is particularly impressive since
final convergence occurred in fall 1967.
which (McNamara observed the following
January) marked a 380 :silo jump from fall
1966. Deviation from the consensus on the
high side went out of style just as it became
objectively most plausible.
Why?
Pressures for conformity in the 1960s
tended to operate against overestimating of-
fense deployment. Overestimating rather
than error had become disreputable. For
example, the Secretary, in January 1964,
stressed that "these longer-range projections
Approved For Release 2008/01/14: CIA-RDP85TO0153R000100030005-3