Assessing technological progress using hazard rate models of R&D communities

HD28

.M414

^3

*<«»*i^ 3f

WORKING

PAPER

ALFRED

P.

SLOAN

SCHOOL

OF

MANAGEMENT

ASSESSING

TECHNOLOGICAL

PROGRESS USING

HAZARD

RATE

MODELS

OF

R&D

COMMUNITIES

Michael

A.

Rappa

MassachusettsInstitute _of Technology

30July1993 Sloan

WP#

3596-93

MASSACHUSETTS

INSTITUTE

OF

TECHNOLOGY

50

MEMORIAL

DRIVE

!j\/lasstuhusettsInstituteofTechnology

ASSESSING

TECHNOLOGICAL

PROGRESS USING

HAZARD

RATE

MODELS

OF

R&D

COMMUNITIES

Michael A.

Rappa

MassachusettsInstitute _ofTechnology

30July1993 Sloan

WP

# 3596-93

1993

MASSACHUSETTS

INSTITUTE OF

TECHNOLOGY

Massachusetts

Institute

ofTechnology

Alfred P. Sloan

School of

Management

50

Memorial

Drive,

E52-538

1

1^.1.1.

ueRA^e

AUG

1

2

1993

Assessing

Technological

Progress

Using Hazard

Rate

Models

of

R&D

Communities*

MichaelA.

Rappa

MassachusettsInstituteof Technolo^

Abstract

This paperdescribes the useof hazard rate models for assessingemerging technologies. It suggests thaton a global scaJe thedecisions ofresearchers, in terms oftheir entrance, continuance, and exit from a field, can serveasan indicatorofthe rate ofprogress in theemergenceofanewtechnology.The approach is

demon-strated using the literature on magnetic bubble memory technology as a source ofdata to determine the contribution-spansofresearchers. Survival and hazard functionsare estimatedbased upon the experienceof

morethan seventeen-hundredresearchers contributingto theheld overtwodecades.

1. Introduction

In thehighly competitive battleground ofmemorytechnology, thebattlefield isstrewnwith

therelicsof conceptsanddevices whosepromiseswere neverfulfilledorhave failedtomeet

theultimatetestoftime (Bobeck, BonyhardandGeusic, 1975).

Around

theworld, each

and

everyday, scientists, engineers, and managers

make

choices

about the direction to

employ

their time

and

resources in developing newly emerging tech-nologies. These are

undoubtedly

difficult judgments to make.

Although

the alternatives

confronting

them

may

be evident, theappropriate direction to pursue is seldom absolutely

clear.

Any

choice is likely to be contingent

on

a

number

offactors that

may

vary in their

clarityand significance over time. In pursuing the development ofa technology, managers

must

be confident in their answers to severalquestions, such as: Will researchersadequately

resolve the technical problems? Will the development ofan alternative technology

prema-turelydiminishits usefulness? Will a sufficientlylarge market

come

about to

make

the cost

of developing the technology

worth

the time and effort?

How

will competitors, suppliers,

and customers react? Finding the answers to such questions is a major preoccupation for

management.

The

factors that

may

influence the rate ofa technology's

development would

appear to

be

many

and diverse.

New

technologies emerge within a

complex

socio-technical system

that typically includes the interaction of

many

individuals

and

organizations, each

making

their

own

decisions about

what

to

do

—

or not to do

—

and

when

(Rappa

and Debackere,

1992;

Van

de

Ven,

1993).

Any

single actor (individual or organization) alone

may

have

relatively little effect

on

the overall process ofemergence. Moreover, factors that appear to

have a major influence

may

have far less actual significance, while factors that appear to

have

minor

consequences, or go relatively unnoticed,

may

havea major effect. Indeed, there

An earlier version of this paper was presented at the

R&D

Management Conference on Technology

Assessment and Forecasting

hdd

atthe Swiss Federal InstituteofTechnology (Zurich, Switzerland), 5-7July, 1993.

2 MichaelA.Rappa

may

be so

many

factors involved that the influence of any single one,

when

considered independently,

may

be highly misleading.

In the faceof such complexity, anyeffort to determineand predict the factors that affect

the rate ofa technology's emergence

may

very well be misplaced.

What

is certain is that a

new

technology emerges through the sustained efforts ofresearchers

who

seek thesolutions

that

make

it viable.

The

decisions of these researchers as a

whole

—

in terms of their

entrance, continuance, and exit

from

a particular technological field

—

may

provide

some

indication oftheir perceptions ofthe rate ofprogress. Thiis, rather than search for specific factors affecting the rate ofa technology's emergence and to decipher their impact, a

differ-ent approach

would

be to ascertain the likelihood ofpersistence

among

researchers

working

in the field.

Within

the

domain

of their education

and

training, researchers normally have

some

degree oflatitude inapplying their skills, and arelikely to pursue the direction they consider

to hold the

most

promise. Researchers are likely to shift their focus

among

different

technologies in accordance with their perceptions

—

aswell asthose oftheir colleagues

and

managers

—

ofthe likelihoodofsuccess and the potential payoff.

While

pursuingan agenda,

researchers

must

periodically reassess their

commitments

based

upon

their senseof

how

well

the

work

isproceeding. Ifthe paceofprogress is reasonablywell, they

may

be

more

inclined

to stickwith it. However, ifprogress is slower than they had anticipated, they

must

decide

to redouble their effortsor ultimately to

move

ina different direction.

It is precisely in this

manner

that the decisions ofresearchers as a

whole

may

signal

changes in atechnology's rate of progress towardcommercialization.

The

aggregate measure of such decisions

may

verywell

embody

all ofthe information availableabout the likelihood

ofsuccessfullydeveloping a technology.

That

is, it

may

represent the cumulative judgments ofall of those

who

work

in the field and

who,

through their decisions to stay with or

dis-continue their activities are signaling their

own

assessment of a technology.

The

basic

assumption is that thereis acorrespondence between the rateof progress in a technical field

and

the likelihoodofresearcherscontinuing in it.

2.

Hazard

Rate

Models

of

Emerging

Technologies

Given

the duration ofactivity for individual researchers in a field, it is possible to

esti-mate

for thispopulation (I) the probability that a researcherwill continue in a field a given

number

ofyears, and (2) the probability that, having been in the field a given

number

of

years, a researcherwill exit the field. In actuarial statistics, these are

known

as the survival

and hazard functions ofa population (Lee, 1992).

The

same

statistics are used in sociology

(Allison, 1984; Yamaguchi, 1992), economics (Lancaster, 1990), andorganizational ecology

(Hannan

and Freeman, 1989).

The

method

wasdeveloped inorderto take into accountthe

indeterminacy of duration fora

segment

ofthe populationwhich remains active at thetime

ofanalysis bycensoring suchcases (see theappendix for theformal derivation ofthesurvival

and hazard functions).

It is posited that shifts in the survival and hazard functions over time

may

serve as an

indicator ofchanges in confidence

among

researchers in a field.

The

survival curve,

illus-trated in Figure 2.1, indicates the cumulative probability ofremaining in the field nyears.

The

implications ofa

downward

shift in the survival function from time period t x.o t+\,

shown

indiagram (a), suggests that the probabilityofremainingin the field nyears has de-creased over time. In the case of

«=5

years, the survival rate declines

from

s^ to s^^y In

AssessingTechrwioguaJProgress

t

Yearsin thefield Yearsinthefield

Figure 2.1: Survivalratecomparisonfromone periodtothenext.

diagram (b) the opposite

outcome

is shown, in

which

the survival function shifts

upward

from one

time period to the next. This suggests an increase in the likelihood ofremaining in

thefield nyears.

Although

the survival function

may

provide a useful understandingofthe likelihood of

continuance in a field, the hazard fijnction can be

more

informative. It suggests the condi-tional probability ofexit from the field, given a researcher hasbeen in the field nyears.

An

increase in the hazard rate

may

signal a lackofprogress in thefield, whileareduction in the

hazard rate

may

suggest that researchers areconfident in thepresent rate ofprogress.

The

hazard curve dynamics are illustrated in Figure 2.2. In diagram (a), the hazard

curveshifts

downward

from time period ^to ^+1. This suggests that the probability ofa

re-searcherleaving the field after oneyeardeclines

from

A, to At ^"'1 ^^"^ probability of leaving

the field afterhaving been init for fiveyears declines

from

h-^ to A4. In diagram (b) the

op-posite result is demonstrated, in which the hazard curve shifts

upward

from one period to

the next. This suggestsan increasein the probability ofa researcher exitingafter one yearin

the field

from

A, to Ai. ^"'l ^'^ increase in the probability ofexiting the field after having

been in itforfiveyears from h-^ to ^4.

Notice that thehazard rate is higher and changes

more

substantiallyforresearchers

who

are in the field only a short time. This implies that as researchers accumulate intellectual

capital in a field they

become

less and less likely relinquish this investmentby switching to

other fields. Thisis consistentwith empiricalstudiesofother fields (e.g.,

Rappa and

Garud,

1992) that

show

that the longera researcher is in a field the less likely that he or she is to leave it. Thus,

we

should expecta higher and

more

volatile hazard rate for researchers with

less time in thefield.

Another

way

to view the change over time in the hazard rate is to plot the rate for researchers with a given duration (d) in the field for each year.

The

hazard rate for d=\ is

MichaelA. Rappa

Yearsin field Yearsinfield

Figure2.2:

Hazard

ratecomparison

from

one periodto thenext.

interesting because itindicates the probabilityofstickingwith it

among

researchers

who

are

new

to the field.

However,

it might also be interesting to see

how much

the hazard rate

changesforresearchers

who

have been inthe field

somewhat

longer (e.g., d='b). Thisformof

analysis is illustrated in Figure 2.3. In diagram (a) the hazard rate is generally decreasing

from year-to-year. This implies that, over time, researchers

who

have been in the field for

both one-year

and

five-years are increasingly

more

likely to remain in the field. In diagram

(b) the hazard rate is increasing from year-to-year for researchers inboth categories,

imply-ing that over timeresearchers in both casesare increasingly

more

likely to leave thefield.

3

Figure2.3:

Change

overtimein thehazardrateforresearchers with one-year

and

five-year

Assessing Tfchrwtogicai Progrtss 5

3.

The

Caseof

Magnetic Bubble Technology

The

historical

development

ofmagnetic bubble

memory

technologyserves to illustrate

the use of hazard rateanalysis for the purpose oftechnology assessment. Magnetic bubbles

are a solid-state integrated circuit technologysimilar to that of metal-oxide semiconductors

in that they are fabricated in a thin epitaxial layer

grown

on a wafer substrate.

The

devices storedata in the form of "bubbles"

moving

in a thin film ofmagnetic material.

The

bubbles

are cylindrical magnetic

domains whose

polarization isoppositeto that ofthe magnetic film

in

which

they are

embedded. Bobeck

and Scovil (1971) found the bubbles to be stable across awide range of conditions and movable from point-to-point athigh speed.

The

initial patents for a magnetic bubble

memory

devicewere filed

by

Bell Laboratories

between

May

and September, 1966. Evidence that interest in the

new

technology grew

rapidlycan be seen in the flurry ofpatents and papersin the late 196C)s

and

early 1970s.

An

extensive bibliography

on

magnetic bubble technology compiled

by

Chang

(1975) of

IBM,

identifies 201 U.S. patents issued (see Figure 3.1) and 1132 papers published between 1969

and mid-1974.

Although

a

number

ofindustrial laboratories

became

active in the

develop-ment

ofmagnetic bubbles during this period, the two firms

most

intensively involved from

thebeginning were

AT&T

and

IBM.

1

%

z

2

Figure

3-1

: Cumulative

number

ofU.S. patents issuedfor magneticbubbletechnology.

Magnetic bubbles were considered to be an important mid-range storage technology

that could fill the niche between high-speed, but high-priced semiconductor

memories

and

low-priced, butlow-speed tapeand disk

memory

devices (Feth, 1976).

Along

with magnetic

bubbles, there were a

number

ofother "gap-filler"

memory

storage technologies under development during the 1970s, such as charged-couple devices

and

fixed-head disks (see

Figure3.2). Magnetic bubble technology wasconsidered an attractive alternativeto existing

magnetic disk and tape systems, in part, because it gready reduced mechanical

movement

(Mees, 1976).

MichaelA. Rappa

100

)L

accesstime(seconds)

high-speedsemiconductortechnologies

W//////^ gapfillertechnologies

WMJMMA

low speeddiskandtapetechnologies

Figure3.2: Price/performance comparison

of

storagetechnologies (Feth, 1976).

In the span of a few years,

magnetic

bubbles had

become

touted in the technical

literature zs an emerging technology withsignificant commercial potential. Predictions for

magnetic bubbles in the early 1970s were uniformly optimistic.

Bobeck

and Scovil (1971)

believed that magnetic-bubble

memories would

be "substantially cheaper than core

memo-riesand up to 10 timesfaster than magnetic-disk

memory

systems

now

widely used for high capacity storage."

The

expectations within the research

community

for magnetic bubble

technologywere echoed in a report by the (U.S.) National

Academy

of Sciences in 1972:

Magnetic bubbletechnologyoffers a radical new waytostoreinformation in thin, transpar-entmagneticcrystaJs and to perform logical operationsby thecontrolled motion of

AssessingTechnologicalProgrfs! 7

diskfilememoriesincomputers andelearoniccentral officesand interface directlywith fast semiconductor memories. They mayproveto beaveryfast,compact,and inexpensiveway

tostoreandprocessdata.

By

1976, it was reported that

some

firms had initiated or expanded pilot production

lines

and

that experts were confident that several commercial devices

would

be introduced

within a year (Torreno, 1976).

The

enthusiasm for magnetic bubble technology

was

attributed to a

number

of factors, including: a decline in the cost of wafer substrates,

making

it competitive with the cost ofsilicon

memory

devices; the discovery of

new

bubble films that could be

more

readily

grown on

wafers; newly developed circuit designs that

enabled the use ofstandard semiconductor manufacturing

equipment

as well as higher storage densities;

and

the development of standard packages similar to those used with

semiconductor

devices.

A

group of leading magnetic bubble researchers

from

AT&T

assessed the future prospects ofthetechnology in the followingmanner:

One

questionthatcanbe askedis "Willbubblessurvive thetestandjointheselectwinner's circle in memory technology?"The present authors,although admittedly stronglybiased, believe they will because they have developed within a relatively short time into an inte-gratedmemorytechnology ofelegant simplicitywhich is nowreadytofillanexistinggapin

the memory cost versus access time spectrum. Most importantly, however, is the future

growth capabilityofthistechnologywhich...will, webelieve, lead to the ultimate

replace-ment for those marvelous mechanical whirling dervishes, the moving head disk and tape

memories. Deeds not words will be the ultimate proof (Bobeck, Bonyhard and Geusic, 1975).

This

optimism

was shared

by

industryexperts

who

anticipated magnetic bubblestorage

densities to increase even

more

rapidlythan semiconductor

memory

technology. At a time

when

semiconductor

memory

deviceswere being produced withdensities in the range of

4-to l6-kilobits, it was reported in

1976

that

one

company

would

begin shipment of

100-kilobitbubble

memory

devices byearly

1977

(Torreno, 1976). It wasalso predicted that

by

1980, there

would

be magnetic bubble devices with 128-kilobitdensities and a 40-percent decline in prices. In comparison, predictions

were

made

for l6-kilobit semiconductor random-access

memories

and similar price decreases over the

same

time period (Feth,

1976).

Although

the future ofmagnetic bubbles certainly appeared bright, by the

mid-1970s

there

were

still a

number

ofissues to resolve before the technologycould gain widespread

acceptance.

As

Cohen

and

Chang

(1975) stated:

Greatprogresshasbeen made inthedevelopmentof magnetic bubbledevicesinthe period

from 1967.... However, thequestion ofwhether bubble memories willeventually be used for largecapacity(>10"bits)computermemorystorage(e.g.,as areplacementformagnetic

diskfiles) isstillopen,and depends onfuture technical progress.

The

cautious tone of their remarks

was

well-founded.

The

high expectations

most

experts held for the commercialization ofmagnetic bubble technology never fully

material-ized.

Numerous

technical problems that researchers sought to resolveproved

more

difficult

than initially estimated.

The

slowing pace of progress in magnetic bubbles

made them

increasingly less attractive than other

memory

storage technologies.

By

the early 1980s,

8 MichaelA. Rappa

advanced at such a rapid pace chat the once perceived "gap" in the storage technology spectrum had closed.

4.

Data

At present, data that could be readily used in the hazard rate analysis of a newly emerging technologyare rarely collected in a comprehensive and systematic fashion. This is understandable, since data on the duration of

commitment

by

individual researchers to the

development ofa technology

would

require a fair

amount

ofeffort to collect and maintain

on a continuous basis over a large

number

ofyears.

Although

data

may

exist in thearchives

of

government

funding agencies and industrial firms, the population ofresearchers actively

working

on

a given technology is typicallysufficiently large

and

widely dispersed to

make

country-by-country,

and

firm-by-firm datacollection impractical.

In thefuture, duration data

may

become

part of the

normal

data collection activities of

government

science

and

technology agencies. Until then,

we

can use a proxy for individual

durationdata: namely, data

on

the span of author contributions to thescientific and

techni-cal literature. Publications in a field represent a fairly well detailed, self-reported archival

record generated

by

researchers in their efixjrt to solve the scientific

and

technical problems

confronting them.

As

a source ofdata, clearly it

would

be very difFicult to

match

the

com-prehensive scope

and

longitudinal nature of the literature using other data collection techniques.

When

taken together, theliteraturecan be viewedasa unique chronologyofthe globalefforts ofresearchers toestablish a

new

field, and can be used todetermine the

persis-tence ofresearchers through an examination of their "contribution-spans": i.e., the time spanbetween a researcher's first and last

known

contribution to theliterature in a field.

Although

data extracted

from

the literature will reflect the idiosyncrasies of publication behavior to

some

degree, the

measurement

ofcontribution-spans is relatively unaffected by

publication frequency. Researchers simply

must

publish at least periodically in order to

establish aspan oftheir activity. In this

manner

contribution-span data can provideuswith

an approximation ofresearchers' persistence in a field. Clearly, in the absence of

more

precise data, the literature is avaluable source ofinformation about thelevel ofactivityand

rateof progress in a field. Indeed, it is not unusual for researchers themselves to see a

con-nection

between

the literature and progress in their field. For example, early in the

emergence of magnetic bubble technology, Bobeck, et al. ₍₁₉₇₁₎ state: "It is evident

from

the

number

of papers concerning all aspects of magnetic bubbles technology that have appeared in the last year that the rate of progress in this comparatively

new

field has

remainedata highlevel."

A

commercially available electronic database (INSPEC) covering the science and

engi-neeringliterature

was

used to identify publications in the field ofmagnetic bubble

memory

devices.

The

database was searched on-line using aset of key terms thatwere

known

to be

commonly

used in the lexicon ofmagnetic bubbles researchers

and

might be either in the

tide, abstract, or classification terms ofa document.

The

computer

search identified over

two-thousand

documents.

The

documents

were retrieved electronically and temporarily

placed in a bibliographic database operating on a personal computer. Each

document

was

carefully inspected in order to ensure the accuracy

and

integrity ofthe search procedure.

Documents

that did not pertain to magnetic bubbles or

were

not research-oriented contributions were discarded. After inspection, the database consisted ofa total of

2025

documents.

AssessingTechnologicalProgress 9

The

documents were

first edited to ensure consistency

among

author and affiliation

names.

The

edited database

was

then used to identify 1734 individuals

who

contributed to

the field over two decades,

from

1969 to 1988.

A

statistical database

was

created consisting

of each researcher, the years they

were

active in the field,

and

the length of their

contribution-span.

The

contribution-span is calculated as the

number

of years that have

elapsed from the first to the last

known

publication for each author. Authors

who

publishin

onlyoneyear aregiven a contribution-span ofoneyear.

The

calculation of contribution-spans is fairly straightforward. Nonetheless, since

authors do not always publish in consecutive years,

some judgment

is necessary to

deter-mine

when

an individual is likely to have ceased active research in the field. Previous

analysisofthe frequency ofyearlygaps between publications for authorswith contribution-spans of

more

than

one

year indicates that the time elapsed without publication is usually

no

more

than three years

(Rappa and

Garud, 1992). Thus, for analytical purposes, all

contribution-spans with gaps of

no

more

than three-years are treated as a contiguous span. In caseswhere therewas a publication gap longer than three-years, theauthoris considered

to have exited and then subsequently re-entered thefield.

5. Results

An

analysis ofauthor contribution spans in thedevelopment ofmagnetic bubblesshows

the rapid growth and decline ofactivityin the field. Figure 5.1 indicates the risein

partici-pation between

1969 and

1976,

when

the field peaked with

426

individuals active.

Subsequent to 1976, the level of participation began to decline

—

remarkably so after

1979

—

such thatby 1988 therewerejusta few dozenauthorsstill contributing to the field.

500

400

300

200

100

Figure 51:

Growth of

themagnetic bubbleresearchcommunity, 1969-88.

About

two-thirds of the authors in the databaseare industrial researchers. Authors

affil-iated with U.S. organizations constitute slightly

more

than halfoftotal population.

The

other

most

represented countries are Japan (19.7%), Great Britain (8.0%), the former

contribu-10 MichaelA. Rappa

cion-spans ofonly one year.

About

10%

have contribution-spans of

more

than five years

and

2.5%

havespans of

more

than ten years.

The

longest contribution-span is seventeen

years.

The mean number

of authorships in a single year is 1.1 (s.d.= 1.0),

and

the largest

number

is 12.

The mean number

ofpublications over an entire contribution span is 3.9

(s.d.=5.4), though

57%

ofall authors havingjust one publication in the field.

The

largest

number

of authorshipsby asingle individual is 57.

The

largest

number

ofauthorsaffiliated

with an organization is 117.

The

first step in the analysis was to

make

a non-parametric estimate of the survival

function for each year from

1974

to 1985.

The

calculations were

made

using the lifetable

approach.

The

results are illustrated in Figure 5.2.

To

simplify the graphics, theyears

1974

and

1978

were selected as representative of the situation before and after the peak year,

1976.

The

survival functions are

shown

in diagram (a).

A

clear shift

upward

of the survival

curve occurred between

1974 and

1978, indicating an increase in the probability of

researcherssuying in thefield. In

1974

the probability ofstaying in the field for five years

was approximately .50. This

compares

with a probability ofabout .75 in 1978. This

con-trasts with the change in the survival curve after 1978 is illiistrated in

diagram

(b).

The

survival curve shifted

downward

such that the likelihood of contributing to the field for five

yearsdeclined from .75 in 1978 to about .53 in 1982.

1 5

ContributionSpan (Years)

5 10

Contribution Span(Years)

Figure 5-2: Changesinthesurvivalfunctionofthemagnetic bubble

community

fi'om (a)

1974

and 1978 and

(b)

1978

and

1982.

Non-parametricestimatesofthe hazard functions werealso calculated using the lifetable

approach.

The

results areillustrated in Figure 53.

The

hazard functions for

1974 and

1978

are

shown

in diagram (a).

Over

the five-year period the hazard function generally shifted

downward.

The

probability of authors ceasing to contribute to the field in 1974, after

having been in the field for five years, was approximately.36. This compares with a

condi-tional probability ofabout .1 1 in 1978.

The

hazard rate for authors with one-year

AssessingTechnologicalProgress 11

The

change in the hazard rate between

1978

and 1982 is illustrated in diagram (b) of Figure 5.3. In the five years subsequent to 1978, the hazard function shifted

upward

for

authorswith one-year contribution-spans, but generally shifted

downward

for authorswith

spans longer than five years.

The

likelihood ofexiting after a one-year contribution-span

more

than doubled from .19 in 1978 to .41 in 1982. However, thehazard ratefor authors with five-year spans is unchanged.

What

is curious are the high hazard rates in

1982

for

authorswith contribution spans of 12-13 years. This

may

be caused the retirement of

more

senior researchers in the field, but this is not necessarily the case. Priorhazard ratestudiesof other fields

do

not support this explanation (e.g.,

Rappa

and

Garud, 1992; Rappa, Debackereand Garud, 1992).

ConcributionSpan (Years)

5 10

Contribution Span (Years)

Figure 5.3: Changesinthehazardfunctionofthemagnetic bubblecommunity

from

(a)

1974

and

1

978 and

(b) 1

978

and

1982.

The

change

m

the hazard rate over time for one-year and five-year contribution-spans {d=\ and d=5) were calculated for the period

from 1974

to 1985 (see Figure 5.4). In the case of authors with one-year contribution spans, the hazard rate varies greatly over time.

Initially it declines from about .35 in

1974-76

to about .20 in 1977-78. After

1977

the

hazard rate increases dramatically from .20 to .50 in 1981. Interestingly, the rate decreases

again to a low of about .25 in 1983, before rising again to near .50 in 1985. Generally speaking, the hazard rate is higher in the post-1978 period. In the case of authors with

five-year contribution-spans, the hazard ratedeclined rapidly from over.35 in

1974

to about .15

in 1975, to a low ofabout .04 in 1977. After 1977, the hazard rate remained fairly stable

from .04 to .1 through 1985.

6.

D

iscussion

Today, the assessment ofemerging technologies is frequentlya matter of sorting-out expert judgment. Technical

and

market experts are asked to provide their best sense of

12 MichaelA. Rappa

Year

Figure 5-4:

Hazard

ratechangesfrom

1974

and

1985

for authors with one-year

and

five-year contribution-spans.

judgment

is

no

simple matter.

A

number

of problems can arise in the interaction

between

experts

and

managers,

and

among

experts themselves, that can affect the accuracy ofthe

judgments they make.

Several techniques have been developed to mitigate the element ofbias inherent in the

solicitation ofexpert opinion

(Kahneman,

Slovic

and

Tversky, 1982; Armstrong, 1985;

Cooke,

1991; Porter, et al., 1992). Nonetheless, a

dilemma

remains in the expert appraisal

of

emerging

technologies. Experts

who

are the

most

knowledgeable about a technology

—

because they are involved in its development

—

are

more

likely to

make

optimistic

predic-tions about its progresstoward commercialization. This isdue, in part, to the difficultysuch

experts have inseparating

what

they

want

to

happen from what

theythinkwill happen.

However,

to resort to experts

who

are less knowledgeable about the technology

—

because they have little actual experience with its

development

—

does little to resolve the

problem.

Not

only might theysimply not

know

enough

to

make

accurate judgments, they

may

also have their

own

commitments

to alternative technologies that blind

them from

seeing the benefits ofthe technology they are asked to appraise.

Much

ofthe technical de-bate thatsurrounds an emergingtechnologyis

embedded

in the conflictwhich arises

among

those experts

who

have a vested interest in seeing it develop, and those experts

who

have a

vested interest thatisdiametricallyopposed.

Theoretically speaking, theemergence ofa

new

technology should not be as difTicult to

comprehend

asitsometimes appears.

The

technologies thatwill be ofthe mostsignificance five-to-ten years from

now

alreadyexist in research laboratories, albeit in a primitive form.

Technologies

do

not emerge instantaneously

from

idea tocommercial application

—

regard-less ofthe impression publicists

may

create. Typically, several years, ifnot several decades,

may

intervene prior to widespread

commercial

use. Technological development is a slow

tech-Assessing TfchnologicaiProgress \3

nologies will be increasing in their

momentum

toward commercial application, while other (perhaps once very promising) technologies are decreasing in their

momentum.

One

ofthe principlechallenges in assessingemerging technologies is to accuratelyascertain, on aglobal

scale, changes in a technology's rate ofprogress toward commercial application.

We

will never

do away

with experts and their judgments about technology.

Beyond

how

we

select experts

and

solicit their opinions, the critical question is whether or not

we

can

improve

upon

the information available to managers (and experts)

when

formulating

judg-ments in technologyassessment. In this regard, hazard models

may

shedadditional light on

the relative rate of global progress in the emergence of

new

technologies. In essence, the

survival

and

hazard functions for a fieldrepresent theaggregatejudgmentsofall the experts, researchers, managers, and others

who may

havea stake in thetechnology's development.

The

results ofthe present empirical analysis,

which

uses contribution-spans as a

surro-gate

measure

for researcher durations in the field of magnetic bubble technology, are suggestive. In the case ofthe survival function, there appears to be a general concurrence with theoretical expectationsof

how

it shouldshift over time.

The

survival function shifted

upward

as the field

expanded

and enthusiasm grew. After 1978, however, the function

shifted

downward,

reflecting thediminished interest in the field.

The

evidence

from

the hazard rate analysis is less clear. Although the hazard rate did

decline substantially for authors with one-year

and

five-year contribution-spans prior to

1978, after

1978

it

was

only in the case of authors with one-year contribution-spans that the hazard ratesubsequently increased. This finding issuggestiveofthe significant influence that accumulated experience in field can have

on

individuals in reducing the likelihood of

their exit

from

a field.

The

empirical findings

must

be interpreted in light ofthe inadequacies ofscientific and

technical literature.

The

omission ofresearchers

who

are either unwilling, uninterested or

unable to publish or present their research is clearly

one

limitation.

While

the analysis is

relatively insensitive to a researcher's publicauon frequency

beyond

some

minimal level to

establish the contribution span, it cannot account for only those individuals

who

do

not contribute to the literature. Ideally, duration data collected

on

a regular and comprehensive

basis

would

be the

most

useful for such an analysis.

Government

and industryshould

seri-ously

examine

the potential usefulness of hazard rate models,

and

the possibility of

collecting duration data for critical pre-commercial technologies as part of the on-going

activitiesto construct

and

disseminatescienceand technologyindicators.

Acknowledgments

The

author isindebted to Professor

Raghu

Garud

(New

York

University) and Dr. Koenraad Debackere

(Gent

University) for their help in developing this avenue of research, and

Professors

Thomas

AJlen

and Edward

Roberts

(MIT)

for theirsupport and encouragement.

References

Allison, P.

D.

(1984) Event History Analysis: _{Regression for} Longitudinal Data, Sage

University Paper Series

on

Quantitative Applications in the Social Sciences, 07-046,

Newbury

Park: Sage Publications.

14 MichaelA. Rappa

Bobeck, A. H., Bonyhard, P. I. and Geusic,

J. E. (1975) "Magnetic bubbles

—

an emerging

new

memory

technology," Proceedings _oftheIEEE, vol. 63, 1 176-1 195.

Bobeck, A. H., Fischer, R. F. and Smith, _J. L. (1971)

"An

overview of magnetic bubble

domains: material-device interface". Proceedings

oftheAIP

17th Conference on Magnetism

and

MagneticMaterials, 45-45 (Reprinted inChang, op cit, 250-260).

Bobeck, A. H., Scovil,

H.

E. D. (1971) "Magnetic bubbles". ScientificAmerican, vol. 224,

78-90 (Reprinted in

Chang,

opctt,119-1A\).

Chang,

H. (1975) Magnetic Bubble Technology: Integrated Circuit Magneticsfor Digital

Storage

and

Processing,

New

York:

IEEE

Press.

Cohen,

M.

S. and

Chang, H.

(1975)

"The

frontiers of magnetic

bubble

technology", Proceeding_oftheIEEE, vol. _63, 1 196-1 206.

Cooke, R.

M.

(1991) Experts in Uncertainty: Opinion

and

Subjective Probability in Science,

New

York:

Oxford

UniversityPress.

Feth, G. (1976) "Memories: smaller, faster, and cheaper"

IEEE

Spectrum, vol. 13.

37-43-Garud, R. and Rappa,

M.

A. (1992)

"On

the persistence ofresearchers in technological de-velopment,"

Academy

of

Management

BestPapersProceedings, iG'^-yi'i.

Hannan,

M.

T.

and

Freeman, _J. (1989) OrganizationalEcology,

Cambridge,

Mass.: Harvard

University Press.

Kahneman,

D., Slovic, P. and Tversky, A. (1982)

Judgment Under

Uncertainty: Heuristics

and

Biases, Cambridge:

Cambridge

University Press.

Lancaster, T. (1990) The EconometricAnalysis _ofTransition Data,

Cambridge:

Cambridge

UniversityPress.

Lee, E. T. (1992) StatisticalMethodsfor Survival

Data

Analysis,

New

York:

John

Wiley

&

Sons.

2nd

Edition.

Mees, C. D. (1976)

"A

comparison

of bubble and disk storage technologies,"

IEEE

TransactionsonMagnetics, vol.

MAG-

12, 1-6.

National

Academy

ofSciences, National Research Council (1972) Physics inPerspective, vol. 2, part A:

"The

coresubfields ofphysics", Washington,

DC,

475-481. (Reprinted in

Chang, ^/)«>., 221-226).

Porter, A. L, Roper, A. T.,

Mason,

T.

W.,

Rossini, F. A., Banks. _J.

and

Wiederholt, B. _J.

(1992) Forecasting

and Management

_ofTechnology,

New

York:

John Wiley

&

Sons.

Rappa,

M.

A. and Debackere, K. (1992) "Technological

communities

and

thediffusion of

knowledge",

R&D

Management,

vol. 22, 209-220.

Rappa,

M.

A., Debackere, K. and Garud, R. (1992) "Technological progress and the

dura-tion of contribution spans", TechnologicalForecasting

and

Social Change, vol. 42,

133-145.

Rappa,

M.

A. and Garud, R. (1992) "Modeling Contribution Spans ofScientists in a Field:

The

CaseofCochlear Implants",

R&D

Management,vol. 22, 337-348.

AssessingTechnologicalProgress 15

Van

de Ven, A. H. (1993) "A

communiry

perspective on the etnergence of innovations,"

Journal ofEngineering

and

Technolo^

Management,

vol. 10, 23-51.

Yamaguchi, K. (1991) EventHistoryAnalysis, Applied Social Research

Methods

Series, vol.

28,

Newbury

Park: SagePublications.

Appendix

Formally, the survival and hazard functions,and the associated probability distribution, are derived as follows (Yamaguchi, 1991): Let 7" _{be a}

_random

_{variable for the duration of} the riskperiod foran event(e.g., exit

from

thefield).

Then

the hazardrateh(f) isgivenas

h(r)

=

lim

P([

+

M>T>t

I

T>t)_

f(0

It

~S(0

where

P{t

+

At

>

T

>

t \

T

>

t) indicates the probability that the event occurs during the

time(t, t

+

Af)given chatthe eventdid notoccur prior totime t.

The

survivorfunction

S(0

isgivenas:

S(0

= ?{T

>t)

=

exp

[

h(«)

du

The

unconditional instantaneous probabilityof havingan event at time t, f(t),

which

isalso

called the probability density functionofT, isgiven as:

P(f

+

Ar

>

7"

_>

t)

fit)

=

lim

—

=

hit)

exp

'-0

_At

I

JAN.

A

lytf^

Date

Due

MIT LIBRARIES OUPl

I I lllll III lllllllllllllllll'11'ffi