DIFFERENCES IN EXPECTED AND ACTUAL
RETIREMENT AGE AMONG OLDER MEN
SCOTT H. BECK
A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL
OF THE UNIVERSITY OF FLORIDA IN
PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR
THE DEGREE OF DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
This dissertation is
lovingly dedicated to my wife
This research, and my graduate education at the University of Florida,
has benefitted from the assistance of many.
First of all, I would like to thank my chairman, Dr. John Henretta,
for his encouragement, sharing of knowledge and, most importantly, for
his patience with me over the past three years. I would also like to
thank him for his aid and encouragement in obtaining outside support for
this dissertation and for his aid in securing the post-doctoral fellowship
that I have received.
This research was supported by Administration of Aging Dissertation
Grant 90 ATO 0 49/01. Also, the support of the Department of Sociology
and the Center for Gerontological Studies has been of considerable impor-
tance in the completion of this dissertation. I would like to give special
thanks to Ray Jones of the University of Florida Library for obtaining the
data tapes used in this study. Data were analyzed using the Statistical
Analysis System (SAS) and the computing was done using the facilities of
the Northeast Regional Data Center of the State University System of
I would like to express appreciation to the other members of my doc-
toral committee. Dr. Gordon F. Streib was primarily responsible for intro-
ducing me to the field of social gerontology and his enthusiasm for all
aspects of research in this area has been of considerable importance in
my graduate education. I would also like to thank Dr. Streib for his
efforts, on my behalf, in obtaining the post-doctoral fellowship at the
Midwest Council for Social Research in Aging. Dr. Benjamin Gorman has
also been of assistance throughout my three years here at the University
of Florida. I would like to express special thanks to Dr. Alan Agresti,
whose comments and criticisms on statistical aspects of this study have
been very helpful. Dr. Agresti is also one of the best teachers I have
ever known and his clear and patient style of presentation has often
helped me to understand otherwise difficult concepts and procedures.
Dr. Cynthia Rexroat has also been a valuable committee member who gracious-
ly agreed to serve on my committee after the dissertation had begun. Other
faculty members not on my committee but who have been of great assistance
throughout my three years here at the University of Florida are Dr. Joseph
Vandiver and Dr. Charles Wood.
I would like to express gratitude to my cohort of sociology graduate
students here at the University of Florida. Their willingness to lend
assistance in academic matters and to be sources of support throughout
the trials and tribulations of graduate school has been of unmeasureable
Finally, I would like to express my debt of gratitude to my family.
My mother has always been a source of strong and consistent support for
whatever I attempted to accomplish and her love is never ceasing. To
Rubye, my wife, I owe the greatest debt of graditude. As a fellow grad-
uate student, she has been supportive and helpful in my academic pursuits
and her love has been the greatest source of strength and motivation
throughout this dissertation.
TABLE OF CONTENTS
ACKNOWLEDGEMENTS. . .
LIST OF TABLES . .
LIST OF FIGURES . .
ABSTRACT . .
CHAPTER I: INTRODUCTION AND CONCEPTUAL
Introduction . .
Statement of Problem .
Conceptual Approach .
The Process of Retirement .
Attitudes, Intentions and Behavi
Adjustment After Retirement. .
Summary . .
CHAPTER II: REVIEW OF LITERATURE .
Timing of Retirement .
Structural Factors .
Health Status . .
Financial Considerations. .
Family Status . .
Attitudes Toward Work .
Attitudes Toward Retirement .
Expected Age of Retirement .
Adjustment to Retirement
Retirement and Life Satisfaction.
Health Status . .
Financial Status .
Social and Demographic Characteristics.
Social Interaction . .
Psychological Correlates. .
CHAPTER III: DATA, MEASUREMENT OF VARIABLES AND
. . iii
METHODS OF ANALYSIS
Data . . . .
Correction for Sample Bias . . .
Measurement of Variables . ..
Factors in the Analysis of Expected and Actual
Age of Retirement . . .
Factors in the Analysis of Adjustment to
Retirement. . . ... ..
Methods of Analysis. . . .
Bivariate Analyses. . . .
OLS Analysis of Expected and Actual Retirement Age. .
Analysis of Adjustment Measures . .
Summary. . . . .
CHAPTER IV: ANALYSIS OF EXPECTED AND ACTUAL AGE OF RETIREMENT.
Introduction . . .
Prediction of Sample-Bias Term . .
Descriptive Statistics of Expected and Actual Age. .
The Relationship Between Expected and Actual Age .
The Stability of Expected Age Over Time. . .
Analysis of Differences in Expected and Actual
Age of Retirement . . .
Analysis of Expected Age in 1966 . .
Analysis of Differences in Expected and Actual Age. ..
Summary. . . . .
CHAPTER V: ADJUSTMENT TO RETIREMENT. .
Introduction . .
Prediction of Sample-Bias Term .
Happiness with Life. .
Analysis of Happiness Measure for
and Not-Retired Men .
Analysis of Happiness Measure for
Evaluation of Retirement Experience.
Summary. . .
CHAPTER VI: SUMMARY AND CONCLUSIONS. .
Introduction . .
The Conceptual Approach. .
NLS Data and Measurement of Factors.
Summary of Findings. .
The Expected and Actual Age .
Adjustment to Retirement. .
. . 152
. . 156
Conclusions . ... .. ... 158
Suggestions for Future Research . ... 165
REFERENCES . . ... . 167
BIOGRAPHICAL SKETCH. . . ... ..... 175
LIST OF TABLES
3-1 Distribution of Original Sample in 1966
by Interview Status. . . ... 44
4-1 Logit Regression for Prediction of Retirement. ... 76
4-2 Descriptive Statistics of Expected and Actual
Retirement Age for Retired and Not-Retired Men 78
4-3 Regression of Actual Age on Expected Age for Full
Group and by Age Cohort. . ... 82
4-4 Regression of Actual Age on Categories of Expected Age 83
4-5 Regression of Expected Age-1976 on Expected Age-1966,
Expected Age-1976 on Expected Age-1971 and Expected
Age-1971 on Expected Age-1966 for Full Group and
Selected Subsamples. . . ... 86
4-6 Regression of Expected Age-1976 on Expected Age-1966,
Expected Age-1976 on Expected Age-1971 and Expected
Age-1971 on Expected Age-1966 by Age Cohort. ... 87
4-7 Standardized Residuals of Independence and Quasi-
Independence Log-Linear Models for Expected
Age-1966 by Expected Age-1976 for Age Cohorts. .... .90
4-8 OLS Regression of Expected Age of Retirement in 1966
for Sample of Retired Men and Selected Subsamples. 95
4-9 OLS Regression of Actual Age of Retirement for Full
Group and Selected Subsamples. . ... 102
4-10 OLS Regression of Actual Age of Retirement with
Cross-Sectional and Change Variables for Full
Group and Selected Subsamples. . ... 109
5-1 Logit Regression for Prediction of Retirement. ... 124
5-2 Logit Regression of Happiness Measure for Full Sample. 127
5-3 Logit Regression of Happiness Measure for Retired Men. 131
5-4 Logit Regression of Pi/P2 Contrast (Negatives Excluded) 138
5-5 Logit Regression of P3/P2 Contrast (Positives Excluded) 141
LIST OF FIGURES
1-1 Atchley's Schema of the Process of Retirement. 7
1-2 Conceptual Model of Factors Involved in Relationship
Between the Expected and Actual Age of Retirement. 17
Abstract of Dissertation Presented to the Graduate Council
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
DIFFERENCES IN EXPECTED AND ACTUAL
RETIREMENT AGE AMONG OLDER MEN
Scott H. Beck
Chairman: Dr. John C. Henretta
Major Department: Sociology
This thesis is a study of retirement that analyzes the discrepancy
between the expected age and the actual age of retirement. The con-
ceptual model used combines Atchley's (1979) general model of the re-
tirement process with the approach of attitude-behavior theory. Three
general factors are hypothesized to determine both the expected age and
actual age. The factors are: 1) constraint factors, 2) job-related
factors and 3) social and psychological factors. A secondary hypothesis
concerns adjustment to retirement. It is hypothesized that discrepancies
between the actual and expected age of retirement, especially earlier-
than-expected retirement, will lead to less successful adjustment to
Panel data collected between 1966 and 1976 on men aged 45-59 in 1966,
were used to investigate these relationships. Because of the truncated
age range of the respondents, the average age of retirement was 61 years,
while the average expected age was about 65 years. A low correlation
was found between expected and actual age. An analysis of change in
expected age over the ten years of the survey, using panel members who
had not retired, showed a large degree of instability in expected age.
In a regression analysis of the expected age among men who had retired,
predictors in all three of the general factors significantly affected
the expected age. Mandatory retirement policies and pension eligibility
reduced the expected age while commitment to work increased the expected
age. Older workers expected to retire later, but this finding may be an
artifact of the data. In the regression analysis of actual age only
work-related health limitations, which reduce the age of retirement, was
significant. In considering the discrepancy between the actual and ex-
pected age, mandatory retirement policies, eligibility for a pension and
higher assets reduced the negative difference between the actual and
expected age, while the existence of a work-related health limitation
and high commitment to work increased the negative discrepancy. With
respect to retirement satisfaction, earlier-than-expected retirement
led to lower satisfaction with retirement.
INTRODUCTION AND CONCEPTUAL APPROACH
The retirement of workers from the labor force has become an increas-
ingly important social issue in the United States. Concern about the
viability of the Social Security system has led to proposed changes in
the age at which workers can receive social security benefits and in the
levels of these benefits. Laws passed by Congress during the past two
decades have reduced the age at which social security benefits are avail-
able and increased the levels of these benefits quite dramatically. The
consequence of these changes, along with macroeconomic changes, is an
underfinanced social security program that is quickly depleting its
financial reserves. Two of the most common types of proposals for rec-
tifying this situation are the reduction in levels of benefits and in-
creasing the age at which workers can receive benefits.
The importance of retirement for social and economic policies has
arisen for a number of reasons. Retirement has not only become more com-
mon among older workers but there has also been a long-term trend toward
earlier ages of retirement (Sheppard, 1976; Walker and Lazer, 1978).
The reduction from 65 to 62 as the minimum age that men can begin re-
ceiving social security benefits, is generally recognized as one of
the most important factors in producing earlier ages of retirement.
Since this change occurred, over half of all new social security
beneficiaries have been under 65 years old (Bixby, 1976). The reduction
or elimination of minimum age requirements for private pensions has
accompanied the age reduction for social security benefits and there-
fore is also considered a factor in the trend toward earlier ages of
retirement (Kleiler, 1978; Atchley, 1980).
The trend toward earlier ages of retirement has been concurrent
with two demographic trends. The proportion of elderly persons has
been increasing in this country and the length of life has also in-
creased (Hauser, 1976). The combination of these demographic changes
with earlier retirement results in a larger proportion of the population
in retirement for longer periods of time. The demographic trends are
not likely to be reversed and therefore patterns of retirement have
been given the most attention by people concerned with the increasing
economic consequences of a greater proportion of retired persons. Nu-
merous solutions have been suggested but the major policy implication
is the need to change governmental and industry policies in order to
promote later retirement and discourage early retirement (Sheppard and
Rix, 1977; Kleiler, 1978).
The timing of retirement is therefore an important issue for re-
search. In this thesis, the timing of retirement will be analyzed in
the framework of the "process of retirement." Some important issues
related to factors that influence the age of retirement will be investi-
gated. For example, the recent liberalization of mandatory retirement
rules was predicated in part on the belief that these rules reduce the
age of retirement for some people. Most social gerontologists doubt
that the liberalization of mandatory retirement will actually result
in later ages of retirement (Smedley, 1979; Atchley, 1980). The effect
of coverage under a mandatory retirement plan on the age of retirement,
controlling for other factors, will be tested in this research by con-
trasting the group of men covered by mandatory retirement rules to the
group of men who are not covered. Furthermore, the impact of mandatory
retirement will be separated from the effect of pension coverage, which
has been posited as one of the major factors in early retirement. A
number of other factors will also be tested for their effects on the
timing of retirement.
As the title of this thesis suggests, the focus of this research
is not retirement timing itself but the relationship between the expected
or planned age of retirement and the actual age of retirement. The rela-
tionship between planned age and actual age of retirement and the factors
that produce discrepancies between these ages is considered to be impor-
tant for a number of reasons. The strength of the relationship between
planned or expected age and actual age of retirement gives an indication
of how accurate individuals' plans, some years before retirement, are
in relation to the actual timing of retirement. This is important
because this relationship, in a general way, indicates whether those
who retired early planned to retire early. At the aggregate level the
degree-of correspondence between the average expected age and the average
actual age may provide an informal estimate of how well trends in the
expected age of retirement foreshadow trends in the actual timing of
The investigation of how external factors may lead to earlier-than-
expected or later-than-expected retirement is also important. In the
context of early retirement, the results of this analysis may give an
indication of how programs or policies can help some workers retire
closer to the age that they planned, and probably preferred, to retire
at rather than dropping out of the labor force early.
The analysis of the timing of retirement in the context of discrep-
ancies between the planned and actual age of retirement is also important
because of possible individual consequences. The adjustment to retire-
ment is one of the most important concerns of social gerontologists
and policy makers in the aging field. Many retirees have some diffi-
culty in adjusting to retirement and the discrepancy between the planned
age and the actual age may be one factor leading to adjustment problems.
Statement of Problem
This research will be primarily concerned with the strength of the
relationship between the expected age and the actual age of retirement
and how external factors may affect this relationship. Therefore, the
research will center around a causal analysis of discrepancies between
the expected and actual age of retirement. This analysis will also re-
veal the important determinants of the expected age and the actual age
Retirement is viewed as a process that encompasses both pre-
retirement and post-retirement phases (Atchley, 1980). This research
will incorporate the full process of retirement by analyzing the effect
of discrepancies between expected and actual age of retirement on ad-
justment to retirement.
In the remainder of this chapter a conceptual model of the process
of retirement is developed. The relevant literature on retirement is
reviewed in Chapter II. The third chapter deals with the data set to be
used in this analysis, the definition of variables and the hypothesized
relationships, and the statistical methods of analysis. The next two
chapters present the results of the analysis of the expected age and
actual age of retirement and of adjustment to retirement. In the
final chapter, the findings are summarized, general conclusions are
drawn and implications of the research are discussed.
There has been relatively little theorizing about retirement timing
and the retirement process. Atchley (1979) has brought together the
ideas of a number of scholars to produce a model of the process of re-
tirement that identifies the important factors that are involved in the
process. This model will be used as a general guide to the interrela-
tionships among social and social psychological factors in the develop-
ment of hypotheses and empirical measures to be used in this research.
Within this general retirement framework, the approach of social psy-
chologists interested in attitude-behavior relationships will be used
to develop the conceptual framework that will guide this research.
Finally, the process of retirement will be extended to include adjust-
ment after retirement. The issues involved in adjustment to retirement
will be discussed in the context of the three main conceptual approaches
in social gerontology: activity, disengagement and continuity theories.
The relevance of the discrepancy between expected and actual age of
retirement for adjustment to retirement is the final topic of discussion.
The Process of Retirement
According to Atchley (1979), the process of retirement involves three
elements: 1) the process by which individuals come to consider retirement,
2) factors that influence the decision to retire and 3) factors that
influence the timing of retirement. Figure 1-1 partially reproduces
the model that Atchley presents. People consider retirement for many
reasons, but primarily because it is a desired goal. Certainly manda-
tory retirement policies, pension plans, health and functional disabil-
ities, layoffs, and peer and employer pressures often play a central
part in the individual's consideration of retirement. In these latter
cases, Atchley maintains that retirement may be considered a result of
external pressures. The decision to retire results from social, eco-
nomic and psychological factors, most of which are shown in Figure 1-1.
Atchley maintains that there are two primary comparisons that
workers consider in deciding the timing of retirement. The first com-
parison is between the worker's financial needs in retirement and his
expected income in retirement. Second, the individual compares his or
her present satisfaction as a worker with what life in retirement is
expected to be like. As shown in Figure 1-1, a number of factors may
affect these two comparisons.
The information the worker has about retirement timing, finances
in retirement and activities after retirement may affect the two types
of comparisons and therefore affect the timing of retirement. The
availability of this kind of information as well as the accuracy of
information will vary widely among workers. Those workers with more
education, those whose company has a retirement planning program and
those covered by pension benefits are more likely to have access to
accurate information about retirement issues.
The individual's personality may also affect the consideration of
these two comparisons. An example given by Atchley is that those persons
L ro L4
X 0 0
* 4- E
4-C *4- --
C U *i-
Ca 4-) -r- 4-
*- a E 4- WJ
4- LS a) -Z 0 0
0 0 S- S- *-
r .C *r- 04-3 4--'
4-' 4-'3 U n
-00 a) 0) 0
L C4--- .- -
.. 4-0 04--
3 C a La
O( S .= C
4- U M 4- -0 0
.*- 3 0cn a
4- 0 C >
4- **- 4-' *r-
* C S-L a
4-) 0) 4-' U
4- r- Cr) S-
< n c
>i 4- 0
0 ro n- 73
E va co 0
4-O O- e
-0 (M4- (o
ns 3 S:
aj 0>1a) 0 L
- S- LO) L 0
L- C 0- D C 43 (>
a) U E rO) aG
nl >, C- ) u
4n-' S- S .) Q 0
Q- GJ Q. 0 I + >..
o E in .3 i- i. e:
E- 0 r* O <0 ( C 3"-3 GJ a) D. L
)3 Q3 "I D. E *r-
( O E 0 0
,/) 1 0
C -- U4
E a) 0-
L UM Z ~
O 0- O
- C >
I /1 .. i
who are active as opposed to passive in relation to their environment
are more likely to plan for retirement. These "planners" are expected
to have more accurate conceptions about financial needs and resources
and also to have more accurate ideas about life in retirement.
The "social/psychological" and "rewards of employment" factors
are posited to affect only the comparison between current situation as
a worker and the expected situation in retirement. The worker's attitude
toward retirement may be especially important in how retirement compares
with current satisfaction as a worker. While job satisfaction and the
meaning of the job to the worker are only applicable to the comparison
of current situation to expectations of life in retirement, financial
rewards of the job may plausibly have an impact on expectations of
financial need and how these needs compare with expected resources
As indicated in Figure 1-1, some trade-offs between the two pri-
mary comparisons can be expected. Workers who want to stop working and
look forward to retirement may forego more adequate retirement income
to retire early. On the other hand, some workers with good financial
prospects for retirement may delay the time of retirement because work
is much more satisfying than the kind of life expected in retirement.
Atchley proposes that if retirement appears more favorable to the in-
dividual than continued employment, the decision to retire will be
made. The two primary comparisons set forth by Atchley may not be
fully considered by some workers. Many workers may only have a general
idea of how much income will be needed in retirement and may not be
sure how much they will receive from social security or the amount of
pension benefits. Also, expectations about retirement may be ambiguous
or not clearly formulated so that for some individuals a comparison be-
tween current situation as a worker and life in retirement is never an
important consideration. Because the two types of comparisons may not
always be considered in the timing of retirement, the set of intervening
factors are also posited to have direct effects on the timing of retire-
One drawback of this model is that it underemphasizes the importance
of the "constraint" factors such as mandatory retirement and health lim-
itations. A worker who knows that the mandatory age is 65 may never
seriously consider needs versus resources or satisfaction with job to
what retirement will be like, but may simply accept the mandatory age
as the time of retirement. Social influences such as marital status
and dependents for whom the worker has financial responsibility are not
explicitly shown as factors but are recognized as potentially important
in the timing of retirement.
The conceptualization of the process of retirement presented above
may be even more applicable to the determination of the expected or
planned age of retirement. If workers do take such factors into account
in the actual decision to retire, then such factors may be considered
some time before retirement when the probable age of retirement is
determined. The main issues in this research involve the relationship
between the expected age and the actual age of retirement and the factors
that are important in explaining the discrepancy between expected and
actual age. A related issue involves differences in the importance of
factors in explaining the expected age and the actual age of retirement.
Because the issues to be addressed in this research are not all en-
compassed within Atchley's model, two approaches to individual behavior
that are usually labelled under "attitude-behavior" theory will be
utilized. By using these more general approaches, it is possible to
develop a conceptual model of the process of retirement that addresses
the primary issues in this research.
Attitudes, Intentions and Behavior
Following LaPiere's (1934) research on discrimination that documented
the large discrepancy that may exist between what people say they will
do and what they actually do, the strength of the relationship between
attitudes or intentions and behavior has been questioned. Psychologists
such as Fishbein and Ajzen (1975) have developed more precise definitions
of "verbal behavior" and more rigorous measures of both verbal behaviors
and actual behavior. Sociologists,on the other hand, have tended to
look for external or intervening factors that may explain behavior, and
consequently explain the discrepancy between attitude and behavior. Both
of these approaches will be discussed and elements of both will be used
to formulate a conceptual model of the retirement process centered
around the relationship between the expected and actual age.
Fishbein (1966) was one of the first to differentiate the concepts
of beliefs, attitudes and behavioral intentions from the general concept
of attitude. These concepts are most definitively presented in Fishbein
and Ajzen (1975). Beliefs are ideas formed around some object (e.g.,
retirement) based on perceived attributes of that object. The totality
of beliefs serves as an information base in the development of attitudes.
An attitude toward some object is an evaluation of the attributes of
that object. Therefore, an attitude by definition encompasses either
positive or negative feelings toward the object. Behavioral intentions
are simply the individual's intention to perform behaviors with respect
to an object. Behavioral intentions are expected to be directly related
to behavior, while beliefs and attitudes are at best indirect determi-
nants. Fishbein and Ajzen conceive the relationships among these con-
cepts as a causal path where beliefs lead to a general attitude that
results in a certain behavioral intention which precedes the actual
One of the main arguments of Fishbein and Ajzen is that an attitude
is only a general predisposition that does not correspond to specific
behaviors but only leads to certain intentions. It should be noted that
intentions and behaviors are only specifically related. That is, some
generalized intention will not necessarily be related to specific behav-
iors. Fishbein and Ajzen explicitly allow for social and other external
influences in their model only through the specification of normative
beliefs concerning a behavior. These normative beliefs result in sub-
jective norms concerning behavior that, in turn, partly determine be-
havioral intentions. The other direct determinant of behavioral inten-
tions is attitude toward the performance of the behavior.
It is apparent that the expected age of retirement is a behavioral
intention while the actual age of retirement is the specific behavior
of interest. In this discussion, therefore, the focus is on Fishbein
and Ajzen's formulation of the relation between intention and behavior.
First, as noted above, the intention and the behavior must be at the
same level of specificity. Second, the intentions and behaviors, if
they involve a single act (e.g., the act of retirement), consist of
four elements: 1) the behavior itself, 2) the target of the behavior,
3) the situation in which the behavior occurs and 4) the time interval
between the two. The third and fourth elements are especially important
in the strength of the relationship between intention and behavior.
According to Fishbein and Ajzen, there should be a high correlation
between an intention and a behavior if three conditions exist. First,
intention and behavior must correspond in level of specificity; second,
the intention should be relatively stable over time; and third, the
carrying out of the intention should be under the person's control. The
expected and actual age of retirement are at the same level of specificity.
The conditions of stability of intention and degree of individual control
over behavior are problematic however. This is where the psychologists
must allow for outside influences in their model.
The stability of an intention is closely related to the amount of
time between the measurement of the intention and the actual behavior.
"The longer the time interval between measurement of intention and ob-
servation of behavior, the greater the probability that the individual
may obtain new information or that certain events will occur which will
change his intention" (Fishbein and Ajzen, 1975:370). The way that
Fishbein and Ajzen handle this "problem" is to greatly restrict the
time interval between statement of intention and the actual behavior
through experimental or quasi-experimental designs.
The degree to which an individual has the ability to carry out
intentions is always problematic. This is bound to be even more true
in natural settings than in experimental situations. In the case of
the timing of retirement, some individuals will have a large degree of
control over their timing of retirement while others will have little
or no control.
For sociologists it is the external influences, especially social
and social psychological factors, that are of greatest interest. Some
sociologists have pursued this matter and formed an approach to the
attitude-behavior relationship that differs from Fishbein and Ajzen's
Much of the impetus for looking at other factors that may be in-
volved in behavior was provided by research on racial attitudes and
actions. DeFleur and Westie (1958) were one of the first to pursue
this and found that most of the discrepancy between what was stated
(liberal attitudes) and later behavior (refusal to engage in integrated
activities) was the result of constraints by reference groups or the
family of the respondents. Fendrich (1967) carried out an experimental
design on racial attitudes, degree of commitment, and behavior. He
found that the situational context of the measurement of attitudes and
behaviors is very important. The situation of an interviewer asking
hypothetical questions is much different from actual situations that
demand an action of some kind.
Warner and DeFleur (1969) present the postulate of contingent
consistency as a framework for attitude-behavior research. This postulate
simply states that intervening variables that modify the relationship be-
tween attitudes (or intentions) and behavior must be taken into account.
In general, this means that the attitudes or intentions of an individual
are only one factor from among a larger set of factors which play a
part in determining the individual's behavior. These intervening fac-
tors are not necessarily occurrences within the period of time between
measurement of attitude or intention and behavior; such influences as
peer or reference groups may have existed before the measurement of the
attitude, but their influence is only exhibited when the action is to
Acock and DeFleur (1972) tested the contingent consistency approach
with data on student's attitudes and behavior toward marijuana. In
analyzing the "behavior" of voting for or against legalization of mari-
juana among college students, the researchers found that perceived
family and peer group support for legalization of marijuana was very
important in whether the student voted "yes" or "no." In this particular
example, it is easy to see how such influences could be included under
the "normative beliefs" factor in Fishbein and Ajzen's model. However,
because Fishbein and Ajzen allow for only an indirect effect of norms
and group pressure on behavior, this effect would be underestimated in
Two sociologists, Albrecht and Carpenter (1976), tested much of
Fishbein and Ajzen's model with the same data that Acock and DeFleur
used above. For the most part, Albrecht and Carpenter support Fishbein
and Ajzen's model, especially the specification of intentions as con-
ceptually distinct from attitude. They warn that the very short inter-
val between measurement of intention and behavior may lead to an un-
realistically high correlation between the two. In most of the
experimental designs, the lapse of time is a number of weeks. This
not only greatly restricts the amount of change in external influences
that may occur but also may cause greater reactivity effects, especially
in non-natural settings. A person who self-reports a behavior that is
directly related to a question on the intention to perform that behavior
asked only a few weeks earlier, is not likely to give a contradictory
There have been a number of sociological studies that test the re-
lation between an intention and later behavior, although the research did
not emanate from either of the above conceptual frameworks. Some of
these studies also looked at other factors that may be considered inter-
vening factors in the relationship between intention and behavior.
Bayer (1969) utilized panel data to analyze the relationship between
expected age and actual age of marriage. Bayer regressed actual age
on expected age, an aptitude measure, parental SES and respondent's
educational aspirations. Expected age was the best predictor but edu-
cational aspirations also had a significant effect. Gasson, Haller and
Sewell (1972) used panel data on high school students to test how edu-
cational and occupational aspirations predict future educational and
occupational attainment. Aspirations were significantly related to
attainment but parent's SES, significant other's aspirations and the
student's aptitude were equally important. Freedman et al. (1965) and
Bumpass and Westoff (1969) looked at expected and desired family size
among women and their later completed family size. There were moderate
correlations between the intended and the actual family size and it did
appear that some intervening changes, such as unemployment of the hus-
band, had an effect on behavior. All of these studies were from
surveys where the time interval was years. The correlations between
intention and behavior were smaller than in Fishbein and Ajzen's re-
search. These lower correlations may actually reflect a more realistic
social psychological process in which intentions to act in a certain
way and the actual action differ because the conditions under which
the two occur are different.
This research centers around the single act of retirement that
occurs at a specific time and in a social context. It is virtually
impossible to set up the experimental or quasi-experimental designs
that are most often used by social psychologists in this type of research
to investigate how intended age relates to actual age. -First, the time
reference is years instead of weeks. Second, although many people may
have some experience with minorities or with marijuana, there is little
"experience" with retirement until it actually occurs. As Tittle and
Hill (1967) point out, the degree of correspondence between attitude or
intention may depend to some degree on the extent that the behavior
constitutes action within the common range of experience. In the case
of retirement, this is usually not true. Because of the time lapse and
lack of experience with retirement, a very strong relationship between
the expected age and the actual age of retirement is not expected. Fish-
bein and Ajzen report correlations of .65 to .85 between intentions and
behaviors. Such high correlations between expected and actual age are
not likely, but some positive relation between these two measures is
The concepts of intention and behavior as derived from Fishbein and
Ajzen, the situational approach of Warner and DeFleur (and others), and
the model of the process of retirement by Atchley are combined to form
the conceptual model that will guide this research. A representation
of the conceptual approach is presented in Figure 1-2.
The constraining, job-related and social and psychological factors
shown in Figure 1-2 are aggregations of Atchley's more detailed model.
All of the specific factors or types of factors that might be important
in the process of retirement will be discussed in the literature review.
These external factors should directly affect the expected age of retire-
ment. The broken arrow between expected and actual age represents a
contingent relationship that may be large or small.
*r- 4- E
> O 0a
.C 3 *r-
a 4-) 4-)
CO U )
The constraining, job-related and social and psychological factors
are shown to influence the actual age, above and beyond their effect on
the expected age. A more appropriate way to think of these external
effects in the context of intention and behavior is that they explain
some of the difference between the expected and actual age.
The external factors may change between the measurement of expected
age and the time of retirement. Therefore, the influences of these ex-
ternal variables may be in the form of changes that the individual exper-
iences, such as worsening of health or change in occupation, that may
help to explain differences between the expected and actual age of re-
tirement. Retirement is, therefore, seen as a process of forming an
expected age of retirement and then of retiring at an age that may or
may not correspond to the expected age. The actual age will often be
different from the expected age since retirement is a dynamic process
where change in the relevant factors, and in the individual's evaluation
of them is likely to occur between the formation of the expected age and
the time of retirement.
Adjustment After Retirement
The process of retirement does not end with retirement itself. The
individual's life is almost certainly changed in a number of ways after
retirement and adjustment to retirement is an important issue. In this
research there is special interest in the relation of the discrepancy
between expected and actual age and certain consequences of retirement
such as lowered income to adjustment of retirees.
It is clear that the major issue in adjustment to retirement is the
individual's adjustment to the loss of a job. There are, of course,
other consequences of retirement such as lower income and lower levels
of social activity. These other consequences do not necessarily occur
and they are usually considered separate factors in the determination of
satisfaction or happiness with life after retirement. Three basic theo-
retical approaches in the field of social gerontology posit different
hypotheses concerning the effects of retirement and these will be
Activity theory is the oldest theoretical approach in gerontology
and was given much impetus by Friedmann and Havighurst's (1954) volume
on work and retirement. In this approach, the job is considered to be
an important and central role for men. In order to adjust to the loss
of the work role, the person must find a substitute that will fill this
void. Therefore, other meaningful activities are necessary for success-
ful adjustment after retirement. In general, retirement is seen as
problematic from this perspective. Shanas (1972) reviewed much of the
retirement literature and concluded that for most retirees the loss of
job is not problematic and in fact most accommodate themselves well to
the loss of their job.
Disengagement theory (Cumming and Henry, 1961) may be considered in
opposition to activity theory. The originators of this theory hold that
retirement is one of the natural occurrences in the process of mutual
withdrawal of the individual and society from one another. This mutual
withdrawal is ultimately based in the biological deterioration that
occurs with age and ends with death. Loss of job is not seen as proble-
matic for the large majority of individuals. This approach is ahistorical
since retirement rarely occurred before this century in this country and
many people today never stop working until they die. Based on their
analysis of longitudinal data on retirees, Streib and Schneider (1971)
proposed a process of differential disengagement. According to these
researchers, disengagement, even from the work role, is neither neces-
sary nor irreversible. People can work part-time and some retirees do
re-enter the labor force. Furthermore, retirement does not necessarily
signal disengagement from other activities or roles, and in fact the
loss of the work role may increase the individual's engagement in
other areas of life.
Continuity theory (Neugarten and Havighurst, 1969; Atchley, 1972)
approaches loss of job in the context of social roles and personality
development over the life cycle. By the time people reach retirement
age their personality, interests, activities and "role set" are relatively
stable. The loss of job will usually be handled by redistributing the
time and energy spent in the job to other roles or activities the person
is involved in. This hypothesis contrasts with the activity theory
assertion that new roles and activities will have to be substituted.
For most people, retirement should not be a cause of maladjustment or
decreased happiness with life, although those persons for whom the job
was very important and who do not have other well-developed roles or
interests, retirement may be a negative experience. Atchley's (1976)
tentative theory of adjustment is a further elaboration of continuity
theory. Atchley proposes that central to the process of adjustment is
the importance of the job in the individual's hierarchy of personal goals.
If job-related goals were unachieved at the time of retirement, there may
be problems adjusting to retirement. The degree of maladjustment will
primarily depend on how important the job-related goals were in the
individual's hierarchy of personal goals. Atchley maintains that most
retirees reorganize their personal goals and drop the importance of a
job from this hierarchy.
Most research indicates that individuals are generally successful
in adjusting to the loss of a job (Atchley, 1976). However, some people
do have problems adjusting to retirement and the reasons for these prob-
lems are not clearly understood. Whether discrepancies between the ex-
pected and actual age of retirement have a negative impact on adjustment
to retirement is of specific interest in this research.
The theories reviewed above do not address the issue of how a dis-
crepancy between the expected and actual age of retirement may affect
adjustment. Atchley's tentative theory does suggest that earlier-than-
expected retirement may have a negative effect on adjustment if the in-
dividual has not achieved a work-related goal. In the Barfield and
Morgan (1969) study of auto workers, it was found that men who planned
to retire early and did so enjoyed retirement somewhat more than those
men who did not plan to retire so early but did. The earlier-than-
expected retirements were usually due to health problems. Price et al.
(1979) categorized a sample of retirees into four groups: "early volun-
tary" (ages 55-61), "early involuntary" (ages 55-61), "on-time voluntary"
(ages 62-65) and "on-time involuntary" (ages 62-65). Those categorized
as early involuntary retirees reported significantly lower satisfaction
with retirement than the other groups. The fact that their retirement
was involuntary suggests that they had to retire before they expected.
These results lead to the hypothesis that discrepancies between the ex-
pected and actual age, especially earlier-than-expected retirement, will
result in a more problematic adjustment to retirement.
The primary purpose of this research is to investigate differences
between the expected age and the actual age of retirement. This re-
search is important for two basic reasons. First, the degree of con-
sistency between the expected and actual age gives some indication of
how accurate individuals' plans some years before retirement are in
relation to the actual timing of retirement. In this context, the
analysis of differences in the expected and actual age may result in
the identification of factors that produce greater consistency in these
two ages. Such an analysis may indicate where planning for retirement
would be more helpful to individuals still in the labor force. Second,
the discrepancy between expected and actual age may have consequences
for adjustment in retirement. Whether such differences do have negative
consequences for individuals will be investigated. Other conditions of
retirement that may affect personal adjustment will also be investigated.
Using a detailed schema of the factors involved in the process of
retirement by Atchley (1979) and the conceptual approaches of Fishbein
and Ajzen (1975) and sociologists such as Warner and DeFleur (1969) in
the analysis of intentions and behavior, a conceptual model was developed
of the factors involved in the determination of the expected and actual
age of retirement and the discrepancy between the two. Because retire-
ment is not a "common occurrence" for most people and the time interval
between measurement of expected age and the time of retirement is years,
a high correlation between the expected and actual age is not likely,
although a positive relationship should exist. Personal characteristics
and situational influences should cause discrepancies between the expected
and actual age. Certain types of influences were posited to be important
in explaining the difference between the expected and actual age of re-
The personal adjustment and satisfaction of persons after retirement
is a part of the process of retirement and some issues and basic theo-
retical approaches were discussed concerning adjustment. While activity
theory suggests that adjustment to loss of job is problematic, the dis-
engagement and continuity theories indicate that for the large majority
of persons the adjustment is successful. The continuity approach, and
to a lesser extent disengagement theory, are supported by empirical re-
search more than activity theory. Of interest in this research is one
particular aspect of the conditions of retirement: the effect of a
discrepancy between the expected and actual age on adjustment to retire-
ment. Other factors related to retirement, such as lowered income, are
of interest because of their possible effect on adjustment to retirement.
In order to specify the important factors in the timing of retirement,
the expected age of retirement and adjustment after retirement, previous
research in these areas will be reviewed in the following chapter.
REVIEW OF LITERATURE
In this chapter, the relevant literature on retirement timing, the
expected age of retirement and adjustment or satisfaction after retire-
ment is reviewed. From this review, the relevant variables for the
analysis of the expected and actual age of retirement and satisfaction
after retirement will be determined. In Chapter III, findings from
previous studies are used to develop hypotheses and models for the
Timing of Retirement
There have been a number of studies that address the issue of the
timing of retirement. The discussion of the findings of these studies
is best organized by the types of factors enumerated by Walker and Price
(1974). The authors first discuss certain structural factors that may
affect general retirement patterns as well as individual retirement
decisions. Walker and Price then group individual factors into five
types: 1) health status, 2) financial considerations, 3) family status,
4) attitudes concerning work and present job and 5) attitudes and expec-
tations concerning retirement.
Included under this general category are governmental and industry
retirement policies and programs, age structure, and inflation and un-
employment rates. Some of these factors do not vary across individuals
and, therefore, cannot explain individual variation. The age structure
impacts on general retirement patterns over time but is not applicable
to the study of individual variation in retirement age. The government
policy on social security benefits varies so little across individuals
that it is not likely to have any measurable effect on individual varia-
tion in retirement timing.
Mandatory retirement policies are applicable only to some groups
of workers and, therefore, may affect individual retirement decisions.
The proportion of workers who were covered by mandatory retirement rules,
before the recent amendment in the Age Discrimination in Employment Act,
is estimated to be between 33 and 40 percent (Reno, 1972; Schulz, 1976).
There have been many polemical arguments against mandatory retirement,
including the argument that it forces workers out of the labor force
sooner than if they had no such restriction. This argument may be true
for some workers but mandatory retirement does not seem to affect most
workers. Schulz (1976) estimates that for a cohort of retired male
workers, only a total of 7 percent were unwillingly retired by mandatory
retirement policies and could not find another job. Using the data to
be analyzed in this research, Parnes et al. (1979) estimate that no more
than 3 percent of men who had retired in their sample of older workers
were forced out by mandatory retirement. Because the retirees in the
sample analyzed by Parnes were disproportionately young retirees, the
3 percent figure probably underestimates the true proportion of workers
affected by mandatory retirement rules. The empirical evidence seems
to indicate that mandatory retirement does not generally reduce the age
Coverage under private pension plans may also be considered a
"structural" factor that varies across individuals. In one of the most
extensive studies of retirement decisions, Barfield and Morgan (1969)
found that the availability of pension benefits at ages earlier than 65
induced many auto-workers to retire early. As briefly noted in the intro-
duction of this thesis, the reduction of ages at which workers can receive
private pension benefits has followed the reduction in the minimum age
that workers are eligible to receive social security. Quinn (1977) found
that eligibility for a private pension was significantly related to the
probability of early retirement among men aged 58 to 63 in the Social
Security Administration's Retirement History Study. Parnes et al. (1975)
report that eligibility for a private pension was positively related to
the probability of retirement for men under 65. It does appear that
pension coverage promotes earlier retirement, mainly because the receipt
of these pensions is usually available before age 65.
Walker and Price (1974) and Sheppard and Rix (1977) maintain that
inflation will delay retirement for some workers. Although rates of
inflation may vary by regional location, this effect would be basically
the same for a cohort of men approaching the retirement decision and,
therefore, cannot explain individual variation in retirement timing.
Also, the indexing of social security benefits may greatly reduce the
delaying effect of inflation on retirement patterns over time. Quinn
(1977) tested the effect of local unemployment rates on the probability
of early retirement and found it to be nonsignificant. Large-scale
layoffs in certain industries may result in earlier retirement of workers
in those industries (Atchley, 1976) but the overall effect of this occur-
rence on retirement timing has not been estimated.
Health status is usually measured as the respondent's subjective
evaluation of his ability to work. Health status has generally been
found to be the strongest predictor of early retirement (Palmore, 1964;
Parnes et al., 1975; Bixby, 1976). There may be problems with a subjec-
tive measure of health status,however, especially when it is based on
retrospective reports. Such reports may be deceptive, or used as a
legitimatory device if the respondent feels that he must present an
excuse for retiring early (Pollman, 1971; Quinn, 1977).
Analyzing the 1969 data from the Social Security Administration's
Retirement History Study, Quinn (1977) found that poor health was the
most important predictor of retirement among men aged 58-63. However,
Quinn also found that there was a significant interaction between health
status and pension eligibility, indicating that men in poor health with
a pension were more likely to retire than men in poor health without a
pension. Other research indicates that those who retire for negative
reasons, such as poor health, are more likely to return to the labor
force than are those who retire for positive reasons, such as receipt
of a private pension or a desire to pursue leisure activities (Stagner,
1979; Price et al., 1979; Motley, 1978).
The consideration of finances in retirement, especially income, is
a very important factor in the decision of when to retire. This assumes
that the individual does not retire because of poor health or some other
limitation on labor force participation. Barfield and Morgan's (1969)
major conclusion was that financial factors are of primary importance
in the timing of retirement. These researchers found that the level of
expected retirement income, the receipt of a private pension, home
equity and expected income from assets were important in the decision
to retire before 65.
As mentioned previously, a number of other researchers have found
private pension coverage to be related to earlier retirement. Quinn
(1977) also found that estimated income from assets was positively related
to the probability of retirement for men aged 58-63. Parnes et al. (1975),
however, found no relationship between level of assets and the probability
of retirement for men under 65. Bixby's (1976) analysis of the Social
Security Retirement History data showed that mortgage on a home delays
retirement. Surprisingly, the earnings level of the most recent job
does not appear to be related to the timing of retirement (Barfield and
Morgan, 1969; Parnes et al., 1975; Bixby, 1976).
The additional financial responsibilities that are often present
when the person is married and has dependents may delay retirement. In
his review of the literature, Sheppard (1976) concludes that married
males remain in the labor force longer. Palmore (1971) found that,
controlling for age, married men worked more weeks than nonmarried men.
Bixby (1976) and Parnes et al. (1975) did not find any relation between
marital status and the probability of retirement for men under 65. The
number of dependents appears to delay the age of retirement for men
(Barfield and Morgan, 1969; Quinn, 1977).
The effect of being married on the timing of retirement for men may
in part depend upon whether the wife also works. The proportion of women
in the labor force has increased dramatically in the last 30 years and
the effect of two-earner families will be increasingly important in the
timing of retirement for men and women. There has not been much empirical
research on this subject. Anderson, Clark and Johnson (1980) used the
Social Security Retirement History data to investigate the reciprocal
effects of husband's and wife's labor force participation. The authors
found that wife's participation in the labor force raised the probability
of the husband's participation, and likewise, the husband's labor force
participation raised the probability of wife's labor force participation.
Attitudes Toward Work
The attitudes of a worker toward his present job, toward work in
general and toward nonwork activities such as leisure and recreational
pursuits, may affect the decision of when to retire. The effect of
these attitudinal variables on the actual timing of retirement has been
tested less than other factors because a panel design is necessary.
Barfield and Morgan (1969) found that workers dissatisfied with
their job were not more likely to have retired early when controlling
for other factors. However, the analysis by Parnes et al. (1975) showed
a significant relation between job dissatisfaction and early retirement.
The strength of commitment to work was also found to have a significant
net effect on the probability of early retirement in the analysis by
Parnes et al. (1975). Barfield and Morgan (1969) found that more spe-
cific attitudes about conditions of the job, such as feelings toward
supervisor and the repetitiveness of the job, did not have significant
net effects on the decision to retire early. Quinn (1977) used objec-
tive measures of job autonomy and hazardous working conditions and found
that these factors were not significant in predicting probability of
retirement for men 58-63.
The occupation of the worker has been found by some researchers to
be important in the timing of retirement. The effect of occupation may
often be linked to specific characteristics or consequences of the type
of work involved. For example, workers in lower skill jobs retire at
earlier ages, but this appears to be due to health limitations that are
the result of the kind of work performed. Some high status workers, such
as professionals, may have the financial means to retire early which is
due to higher lifetime earnings that result in greater wealth. Using
weeks out of the labor force as a measure of retirement, Palmore (1971)
found that those in higher status occupations retired later than other
workers. This effect may be due to the greater availability of part-
time work in many upper status occupations. Rubin (1973) analyzed those
persons receiving late entitlement to social security benefits (age 66
or later) and found that persons in lower skill occupations without pri-
vate pension coverage were overrepresented. White collar workers and
professionals were underrepresented in this group, probably because
they continued working full or part-time. Using a nine-category occupa-
tional classification, Bixby (1976) found that those in blue collar
occupations were more likely to have retired early than those in other
occupational groups. Reno (1971) found that those retiring at 65
tended to be in skilled blue collar or lower white collar occupations
and were covered by compulsory retirement and private pension plans.
In general, when controlling for factors such as health and private
pension coverage, there may be no differences in retirement timing
among occupational groups. The exception may be upper-status workers
such as professionals and businessmen who continue working at later ages.
Attitudes Toward Retirement
Atchley (1979) and Walker and Price (1974) maintain that those more
favorable to retirement will retire at earlier ages. As with attitudes
toward work, there has been relatively little empirical research on the
effects of attitudes toward retirement on the timing of retirement be-
cause of the necessity of a panel design. Bixby (1976) used general
attitude toward retirement in 1969 in her analysis of retirement over
the period 1969-1973. Those more positive towards retirement were found
to be significantly more likely to retire than those less favorable to
retirement. Barfield and Morgan (1969) did not have a direct measure of
attitudes toward retirement but did measure leisure and activity plans
after retirement. When controlling for other factors, those men plan-
ning to engage in hobbies or in charitable work of some kind were not
more likely to have retired early. It seems reasonable that persons
who view retirement favorably are more likely to retire at earlier ages
but further research is necessary to determine how important such an
attitude is in the decision of when to retire.
In summary, it appears that health status is the major determinant
of the age of retirement. Financial considerations such as the receipt
of a private pension and other unearned income are important in the tim-
ing of retirement. Family status may be important in the decision of
when to retire. The effects of attitudes toward work, toward the specific
job and toward retirement on retirement timing may be important but ad-
ditional longitudinal research is needed in these areas.
Expected Age of Retirement
Those factors that best explain the age of retirement, health and
financial status, are also important in explaining differences in the
expected age of retirement. Because of this complementarity, separate
sections on types of factors will not be used here.
Poor health does appear to result in younger expected or planned
ages of retirement (Barfield and Morgan, 1969; Barfield, 1970). Ekerdt's
(1979) analysis of preferred age, which is highly correlated with expected
age, showed that health status was related to the preferred age.
In the Barfield and Morgan (1969) study, financial factors were
important in the planned age of retirement. The level of expected retire-
ment income, the receipt of a private pension and assets were all related
to planning early retirement. Ekerdt (1979) also found that retirement
finances were related to the preferred age of retirement. In a later
survey, Barfield and Morgan (1978) report that pension eligibility re-
sulted in lower planned ages of retirement. In this later survey,
Barfield and Morgan also found that having mortgage payments resulted
in later planned ages.
The level of job satisfaction was found to be related to the expec-
tation of retiring before age 65 in the Barfield and Morgan study (1969).
Rose and Mogey (1972), however, did not find a significant net effect of
job satisfaction on the preferred age of retirement. Negative attitudes
toward present job and toward work were found to significantly reduce
the preferred age by Ekerdt (1979). In his analysis of the preferred
age, Ekerdt also found that workers looking forward to leisure activities
tended to prefer younger ages of retirement.
Rose and Mogey (1972) found both education level and occupational
status to be positively related to the preferred age of retirement.
Goudy, Powers, Keith and Reger (1980) found self-employed professionals
to have the highest mean age for "best age of retirement," with other
professionals and proprietors slightly lower and blue collar workers
with the lowest mean age. The authors also found that best age of re-
tirement was related to an "avoidance of retirement" scale, indicating
negative attitudes toward retirement lead to higher preferred ages of
retirement. Neither Rose and Mogey (1972) nor Goudy et al. (1980) take
into account coverage under mandatory retirement plans, which may reduce
the expected or preferred age of retirement.
One of the most important predictors of expected or preferred age
of retirement is age. Barfield and Morgan (1978) report that the "50-54"
and "55-59" cohorts were most likely to plan to retire early, while the
"60-64" cohort and cohorts under 50 had somewhat higher planned ages.
Because of this nonlinear relationship, Barfield and Morgan maintain
that this difference is a cohort effect and not due to aging. Rose and
Mogey (1972) also analyzed cross-sectional data and found that age was
positively related to preferred age. These authors explain this find-
ing primarily in terms of an aging effect wherein workers getting closer
to retirement delay the preferred age of retirement. Ekerdt, Bosse and
Mogey (1980) analyzed panel data and show that as individuals age, they
tend to change toward later preferred ages. Barb's (1977) analysis of a
panel of Iowa men also indicates that as respondents age, the'direction
of change in preferred age tends to be toward later ages. There prob-
ably are some differences among cohorts in the expected or preferred
age, just as there have been differences in the age of retirement among
cohorts. However, most of the difference found to exist between cohorts
in the expected or preferred age appears to be due to aging.
The relatively sparse research on the expected age of retirement
generally shows the same types of factors that are important in the tim-
ing of retirement are also important in the determination of the expected
age of retirement. In analyzing the expected and the actual age, the
same predictors will be used. It is clear that cohort must be taken into
account in the analysis of the expected age of retirement.
Adjustment to Retirement
Part of the retirement process entails what happens to the person
after retirement. One of the major interests in gerontological research
has been the study of factors that are important in determining adjust-
ment to retirement. In this research, investigation of the retirement
process is extended to include the analysis of adjustment to retirement.
The two primary questions in this analysis are whether differences exist
in the satisfaction or happiness of retirees and workers; and among re-
tirees, whether discrepancies between the actual and expected age affect
The review of previous research will begin with a discussion of the
effect of retirement on life satisfaction. Factors related to the satis-
faction or happiness of older people will be reviewed in the following
areas: 1) health status, 2) financial status, 3) social and demographic
characteristics, 4) social interaction and 5) psychological correlates.
Retirement and Life Satisfaction
Perhaps the most often studied aspect of retirement is the satisfac-
tion or happiness with life after retirement. The major issue in this
area has been whether the loss of the work role results in lowered satis-
faction. In one of the first longitudinal studies of retirement, Thompson,
Streib and Kosa (1960) investigated the impact of retirement on life sat-
isfaction. Using life satisfaction scores on respondents when all were
in the labor force and their scores years later when some of the respon-
dents had retired and others were still working, the authors found that
retirees did have somewhat lower scores. The authors analyzed this dif-
ference between retirees and workers by looking at a number of other
factors besides employment status. Lower income and poor health as well
as pre-retirement attitudes were found to affect life satisfaction, and
when one or more of these factors were controlled, no significant differ-
ence between workers and retirees was found.
Using cross-sectional data, Edwards and Klemmack (1973) did not find
any significant differences in satisfaction with life between retirees
and workers when controlling for other factors. Palmore and Luikart
(1972) also used cross-sectional data and did find that employed persons
were more satisfied with their life than retirees, although this differ-
ence was small. In general, the traditional hypothesis that the loss
of the work role itself is harmful to psychological well-being has been
discounted. The issue has not been settled however, and in this research
the possible effect of the loss of the work role will be tested.
The research reviewed in the section on timing of retirement indi-
cates that health may be the most important determinant of age of re-
tirement. Health status also appears to be the most consistent and
strongest predictor of satisfaction or happiness with life (Barfield
and Morgan, 1978; Chatfield, 1977; Edwards and Klemmack, 1973; Bultena
and Oyler, 1971; Medley, 1976; Palmore and Luikart, 1972; Sauer, 1977;
Spreitzer and Snyder, 1974). It is no surprise that health is very
important since health problems can restrict the activities of a person
as well as affecting how they feel.
Palmore and Luikart (1972) were able to compare the effects of sub-
jective health status, which is the usual measure of health status, with
physicians' ratings of the respondent's health. The authors found that
subjective health status was more strongly related to life satisfaction
than ratings of the physicians. This seems to indicate that it is how
people feel about their health as opposed to their actual health condi-
tion that impinges on overall satisfaction.
The standard of living has generally been recognized as important
in the enjoyment of life and psychological well-being of people in all
age groups. Income has been found to have a significant positive effect
on satisfaction, although the size of this effect varies by study (Bar-
field and Morgan, 1978; Elwell and Maltbie-Crannel, 1981; Edwards and
Klemmack, 1973; Palmore and Luikart, 1972; Hutchinson, 1975; Spreitzer
and Snyder, 1974).
The relative level of income may be important for satisfaction
after retirement. Streib and Schneider (1971) show that the reduction
in income after retirement was a major factor in the reduction of life
satisfaction. Barfield (1970) reports that the ratio of retirement income
to pre-retirement income was positively related to satisfaction among
auto workers. In a national probability sample, Barfield and Morgan
(1978) did not find the ratio of retirement income to pre-retirement
income to have an effect on satisfaction, net of present income level.
Tissue (1970) found that middle-class elderly who had fallen into poverty
(defined by eligibility for old age assistance) were significantly less
satisfied with their life than were working-class persons who were also
poor. It may be that only steep drops in level of income produce a
lowering of satisfaction with life.
Social and Demographic Characteristics
Occupational status has often been found to have a positive effect
on life satisfaction or morale, even among retirees (George and Maddox,
1977; Alston and Dudley, 1973). Edwards and Klemmack (1973) found a
significant effect for occupation even when controlling for income. Edu-
cation level usually does not have its expected positive effect (Spreitzer
and Snyder, 1974; Edwards and Klemmack, 1973; Elwell and Maltbie-Crannel,
1981). In fact, George and Maddox (1977) found a slight negative effect
for education when controlling for occupation.
The loss of a spouse, either through death or divorce, appears to
have a detrimental effect on satisfaction or morale. Married persons
have been found to be more satisfied by some researchers (Hutchinson,
1975; Edwards and Klemmack, 1973; George and Maddox, 1977). However,
other researchers have not found any significant difference between
married and nonmarried persons (Barfield and Morgan, 1978; Palmore and
Luikart, 1972). These differences in results may be due to fluctuations
in the distributions of the types of nonmarried persons. Those who are
widowed may be less happy than those who are divorced or separated, who
in turn may be less happy than never-married persons. Because these
types of marital statuses may constitute different proportions of the
general "nonmarried" category, the specific statuses should be used
Two factors that do not appear to have any effect on life satisfac-
tion among older people are age (Palmore and Luikart, 1972; Spreitzer
and Snyder, 1974) and sex (Edwards and Klemmack, 1973; Hutchinson, 1975;
Palmore and Luikart, 1972). Spreitzer and Snyder (1974) did not find
significant differences in satisfaction between blacks and whites;
however, Sauer's (1977) analysis by race of lower-class elderly in
Philadelphia did uncover different predictors of satisfaction for blacks
The degree and type of interaction with others have been regarded
as important factors in the satisfaction or happiness of older persons.
Most social interaction variables have been found at one time or another
to be significant predictors of satisfaction or happiness. Edwards and
Klemmack (1973) found that frequency of interaction with neighbors, fre-
quency of phone contacts with friends and relatives, and the number of
friendly neighbors all increased life satisfaction. These researchers
also report that participation in formal organizations, frequency of
contact with non-nuclear kin and the frequency of contact with offspring
not in the home had no effect on satisfaction. Spreitzer and Snyder
(1974) report that church attendance had no effect on satisfaction.
Palmore and Luikart (1972) report that the number of social activities
was a positive predictor of life satisfaction and also report that par-
ticipation in formal organizations increased the level of satisfaction,
a finding that contrasts with Edwards and Klemmack. Bultena and Oyler
(1971) found that greater social interaction was positively related to
satisfaction but the general measure of interaction used by Conner and
Powers (1975) was not significant. Elwell and Maltbie-Crannel (1981)
report that the degree of informal socializing and the participation in
formal organizations increases the level of satisfaction while family
interaction does not. The inconsistency in the above findings may be
due to differences in the measures used.
In a review of the literature on correlates of life satisfaction,
Adams (1971) reports that the self-identidication as old by the elderly
is related to lower levels of satisfaction. Adams also finds that feel-
ings of inadequacy by older men and feelings of rejection by older women
are generally correlated with lower satisfaction. In the social-
psychological sphere, Adams reports that most research has found that
the "contraction of life space" leads to lower satisfaction. This lat-
ter finding is in direct contrast to the disengagement hypothesis of
Cumming and Henry (1961) who argued that disengagement, which includes
contraction of life space, is part of the normal process of growing old
and does not result in lowered happiness or satisfaction with life. One
psychological variable that Palmore and Luikart (1972) found to be sig-
nificant in predicting life satisfaction was internal-external locus
of control. This is a measure of the extent to which an individual
believes he has control over his own life. Those who did believe that
they have a great deal of control over their own lives (inner-directed)
were significantly more satisfied.
In summary, subjective health status is the most important determin-
ant of satisfaction with life. One aspect of this relationship that has
apparently not been tested is the effect of length of present health con-
dition on present satisfaction. It may be that those persons who have
chronic conditions or impairments for a number of years come to accept
and cope with such conditions and, therefore, are relatively more sat-
isfied than persons who have recently undergone a worsening of their
health. The standard of living appears to be important in the satisfac-
tion or happiness of older people. Occupational status appears to be
positively related to satisfaction while education and age and sex do
not seem to have an effect on levels of satisfaction. The impact of
retirement appears to work primarily through health status and income
levels, and there is not much empirical evidence to support the "role
loss" hypothesis concerning retirement. Marital status may have an
effect on satisfaction but more specific categories are needed to ade-
quately assess the impact of widowhood and divorce. Although inter-
action with others does seem to play a part in personal satisfaction
with fife, it is certainly not clear from the research which types of
social interaction make a difference in levels of satisfaction. One
factor that has not been previously tested is the effect of the
discrepancy between the expected and the actual age of retirement. As
stated in the previous chapter, it is expected that the larger the dis-
crepancy, especially negative discrepancies, the lower the level of
satisfaction or happiness. The specific relationships to be tested
in this research are stated in the following chapter.
DATA, MEASUREMENT OF VARIABLES AND
METHODS OF ANALYSIS
In this chapter, the data to be analyzed will be described; methodo-
logical considerations will be discussed; the measurement of variables
will be enumerated; and the statistical methods of analysis will be pre-
The data to be utilized in this research are from the National
Longitudinal Surveys of Mature Men, aged 45-59 in 1966. These longitu-
dinal surveys have been designed by Ohio State University for the Depart-
ment of Labor which commissioned the surveys (see Parnes et al., 1979).
The interviewing has been conducted by the Census Bureau. This national
probability sample is stratified by race in order to insure adequate
numbers of blacks for analysis. Thirty percent of the sample is black.
The data cover a period of ten years, 1966-1976, in which respondents
were interviewed eight times. Five of these surveys were face-to-face
interviews (1966, 1967, 1969, 1971 and 1976). In 1968 a mail-out
questionnaire was used, while in 1973 and 1975 telephone interviews
were conducted. Because the 1968 questionnaire contains so little of
the necessary information for this analysis, it is not used here.
The original sample size of approximately 5,000 men in 1966 dropped
to slightly under 3,500 in 1976. Of the 1,500 who were lost, over half
died during this ten year interval. The refusal rate has always been
low; cumulatively over the ten years less than 14 percent of the original
sample refused or could not be located for two consecutive waves. The
interview status of the original respondents and the labor force status
of respondents interviewed in 1976 are shown in Table 3-1. The loss of
respondents, whether by death or refusal, could cause bias in the char-
acteristics of the respondents who remain in the study. However, it does
not appear that any substantial variations in attrition of men by racial,
occupational or income characteristics occurred. Parnes et al. state,
"It seems fair to conclude that attrition has not departed sufficiently
from a random pattern to constitute a serious problem" (1979:9).
The National Longitudinal Surveys (NLS) data contain information on
labor force participation, income and financial matters, family structure,
health, attitudes toward work, and other areas of an individual's life.
The data contain the necessary information on whether the respondent is
retired. The age of retirement was never ascertained in this data set
and, therefore, it must be estimated with the use of longitudinal infor-
mation for each individual.
The decision concerning whether the respondent is retired is condi-
tional on two responses. In every wave a question was asked concerning
the labor force status of respondent. One of the responses to this
question is "retired." In most waves, those persons who responded any-
thing other than retired for their labor force activity were asked,
"At what age do you expect to retire from a (your) regular job?" One
of the responses was "already retired." Most of those responding
Distribution of Original Sample in 1966 by Interview Status
Status (1976) Frequency Percentage
Deceased 841 16.7
Dropped-out 692 13.8
Retired in 1966 139 2.8
Retired 1967-1976 1,246 24.8
In labor force in 1976 2,102 41.9
TOTAL 5,020 100.0
"already retired" were "unable to work" in the labor force activity
variable. Very few men who were "working," "unemployed" or "looking
for work" in the labor force activity variable responded that they were
retired in the question on expected age of retirement. Those men who
were in the labor force but responded that they were "already retired"
on the expected age of retirement question are considered to be retired.
The large majority of men classified as retired were not in the labor
The first step in determining age of retirement is determining the
year of retirement. For those retired in 1976, this involves "stepping
back" one wave at a time until the respondent can no longer be classified
as retired. For example, a man may be retired in the 1976 and 1975 waves
but not in the 1973 wave. For some respondents, determining the year and
month of retirement was facilitated by questions in 1976, 1971 and 1969
that asked the year and month the person last held a regular job. If the
year given was within the time period the person went from not being re-
tired to being retired, that information was used to specify the respon-
dent's age of retirement. For some respondents, there was no information
on last regular job; or their last reported regular job preceded the last
survey year in which they reported themselves as being in the labor force;
or they gave a year for their last job that was later than the first sur-
vey that they reported themselves as retired. Those respondents for whom
there was no information on last job held, or for whom last job held
preceded the last survey year in which they were officially still in the
labor force, the midpoint between the survey year in which they were not
retired and the survey year in which they first reported being retired,
was used as the estimated time of retirement. The midpoint between the
1973 and 1975 surveys is July, 1974; the midpoint between the 1971 and
1973 surveys is July, 1972; the midpoint between the 1969 and 1971 sur-
veys is July, 1970; and the midpoint between the 1967 and 1969 surveys
is July, 1968. The reason for specifying the month of July is that
interviewing took place between the months of June and September, with
most interviews completed in July. For those few men who reported that
the last year they held a regular job was after the year they first re-
ported being retired, a determination of whether this job was full-time
was made (35 or more hours per week and more than 26 weeks a year),
based on information from the next survey wave. If a man had retired,
gone back to work full-time, and retired again within the span of two
survey waves, the latter date of retirement (month and year of last reg-
ular job) was used. Once the month and year of retirement was determined,
the month and year of birth of the respondent was used to determine his
estimated age of retirement. Using this method, it is unlikely that the
error in estimation of age of retirement is more than one year.
Based on these decision rules, a total of 1,246 retirees were iden-
tified. Most of these men retired from a full-time job and had no labor
force experience after retirement. There were very few men who retired,
returned to the labor force, and retired again. Only 84 of the 1,246
retirees (6.7%) worked part-time (less than 35 hours per week or less
than 26 weeks per year) after they had retired from a full-time job.
Most of these 84 men worked fewer than 20 hours per week and fewer than
20 weeks out of the year. Therefore, almost all of the 1,246 men iden-
tified as retired were not working, and those who were working identified
themselves as retired.
Correction for Sample Bias
It is the purpose of this research to estimate parameters of differ-
ent models that should apply to all members of the age cohort who were
45-59 in 1966. However, if the sample of retirees is nonrandom, the co-
efficients obtained will confound structural parameters with the deter-
minants of the probability of sample selection since those who are retired
may be different in unmeasured ways from respondents who are observation-
ally the same (Rexroat, 1980). Because the analysis of differences in
expected and actual age of retirement splits the representative sample
of men in the NLS survey into those who have retired and those who have
not, it is very likely that this sample selection rule (retired vs. not-
retired) results in a group of retirees in which some have a "propensity
to retire" that may bias the estimated parameters. Heckman (1980) sug-
gests treating bias as a specification error in the estimation of coeffi-
cients. His approach will be used to correct for this problem among the
"subsamples" of retired and not-retired men.
To illustrate his approach, an equation is presented in which the
age of retirement is posited to be a function of a set of predictor var-
iables, as well as a disturbance term;
Y = x'B + e, (1)
where x is a vector of exogenous regressors and B is the vector of param-
eters attached to the x's, and e is the error term in which the usual
OLS restriction that
E(e) = 0 (2)
is assumed to be true.
Heckman (1980) argues that in the case where respondents are missing
on the dependent variable, there is the possibility that the conditional
mean of the error term is not zero and that this nonrandom distribution
may be related to the selection rule (retired only) that causes some
cases to be eliminated from the analysis. The actual form of the equa-
tion (1) may then be;
Y = x'B + (x + u), (3)
where x and u are two separate components of the disturbance e. u is the
random error term that is not correlated with any of the x's. x on the
other hand, is a component that involves the person's propensity to be
retired; that is, to be in the sample. The pertinent question is whether
this x is in any way related to the x's in the regression equation, there-
fore biasing the coefficients. Most researchers have assumed away this
possible effect, or qualified their findings by verbally announcing that
there may be some selection bias. Heckman argues that there will usually
be some relation between x and the x's, especially when the same type of
factors that affect the dependent variable are also related to the prob-
ability of sample inclusion. Therefore, the estimate of x is a measure
of certain unmeasured characteristics that bring the person into the
sample; in this case, bring them into retirement. Heckman asserts that
the estimation of a sample bias term that is included in the regressions
should control for the effects of unmeasured characteristics on the
parameters estimated for the independent variables.
The problem is thus changed from one of missing cases on the de-
pendent variable, which is nonresolvable, to one of a missing exogenous
variable, x, in the regression equation. Based on the assumptions de-
veloped above, the estimate for x is in fact an estimate of the covari-
ance of the errors in the equation that predicts sample selection and
the errors in the equation that predicts age of retirement (Fligstein
and Wolf, 1978).
Though most researchers who have used this technique have derived
A from a probit equation, it is also possible to obtain it from a logit
model. The parameter estimate of x is a function of the predicted
probability of sample inclusion; in this case, the predicted probabil-
ity of being retired. The predicted probability can be obtained from
a logit analysis. This probability of being retired can be expressed
Pr = Prob(Y = 1) = F(xB), (4)
where F(xB) is the cumulative distribution function describing the rela-
tionship of the probabilities to the exogenous variables (see Hanushek
and Jackson, 1977). The logistic distribution is defined as Pr
(l+X where the predicted cumulative probabilities range from 0 to 1
as xB ranges from -- to +m. The predicted probabilities can then be
standardized to the normal distribution by the following equation:
S(Pi) = ir ( 5)
where Pir is the predicted probability of retirement for the ith individ-
ual, u is the overall mean of the predicted probability of inclusion and
o is the standard deviation. These standardized probabilities, which are
Z scores, are used to estimate the area under the standard normal distri-
bution which is to the left of the standardized probability, and is
denoted as T. The value for the ith individual is the denominator of
x. The numerator is the ordinate (height) of the standardized probability
on the standard normal distribution at SPir and is expressed as
1 *[-.5(SP )21
Vir= (1) e [-(SPir)2 (6)
where SPir is the standardized probability for the ith individual. As
Heckman indicates, X is the ratio of the ordinate to the tail area of
the standard normal distribution. The formula for the bias-correction
estimate is therefore,
A- r (7)
The values of x are a monotone decreasing function of the probability of
sample selection. Large values of x indicate a greater degree of "biased-
ness" and, therefore, reflect a lower probability of sample inclusion.
An example is given in order to more clearly show how the x term
may correct for bias. Consider the relationship between the actual age
and the expected age of retirement, which are hypothesized to be positive-
ly correlated. Higher values of expected age should result in lower
probabilities of sample inclusion, and consequently higher values of X.
That is, persons who are not very likely to be retired, but nonetheless
are retired, have a high A. Those persons with a high unmeasured "pro-
pensity to retire" (high values of x) will usually be young retirees and,
therefore, a negative relationship between the age of retirement and the
bias-correction term should exist. Using the well-known formula for a
partial correlation controlling for a third variable (Agresti and Agresti,
1979), possible correlations are entered to show the results of the inter-
relations posited above, y is the age of retirement, x is the expected
age and v is the estimate of A.
r v- ryx(ryv)(rxv .30-(-.40)(.15) .40.
yx-v [(l-r2 )(-r2xv)] [(1-.16)(1-.05)]12
The result indicates that if the "true" correlation between actual
age and expected age is .40, in a biased sample it will be reduced to .30
given the negative relationship between actual age and the bias-correction
term and the positive relationship between expected age and the bias-
correction term. This bias-correction term will be added to the basic
regression model (1) which can now be written as
(1.a) Y = x' + X + e.
This equation should control for the potential bias that is due to the
sample selection rule. The exogenous variables to be used in obtaining
the predicted probabilities will be enumerated later in the chapter.
Measurement of Variables
The variables to be used, and the way in which they are measured,
will be explained in this section. In the first section, the variables
related to the analysis of the actual and expected age of retirement are
presented in the context of the conceptual model presented in Chapter I.
In the second section, those variables related to the analysis of adjust-
ment to retirement are presented.
Factors in the Analysis of Expected and Actual Age of Retirement
The dependent variables, expected and actual age, are discussed
first. The important control of cohort is then briefly discussed. The
predictor variables are presented in the context of the three general
types of factors presented in the conceptual model in Chapter I; con-
straining, job-related and social and psychological factors.
Actual age of retirement. The construction of this variable is
discussed extensively earlier in the chapter. The age of retirement
ranges from 47 to 70, even though the upper age for men in this sample
is 69. This occurs because some of those men who were 59 in 1966 were
interviewed shortly before their 60th birthday and in 1976 were inter-
viewed shortly after their 70th birthday.
Expected age of retirement. This variable is based on the question,
"At what age do you expect to retire from a (your) regular job?" The
responses to this question include actual age estimates as well as "al-
ready retired," "never retire" and "don't know." This question was asked
in every wave and the responses in 1966 will be used to estimate the re-
lationship between the expected and the actual age of retirement. Those
men who were retired in 1966, approximately five percent of the original
sample, are deleted. Those responding "never retire" are, for some analy-
ses, assigned an expected age of retirement based on life expectancy by
age and race. The values of these expected ages for the "never retire"
group range from 74 to 78. Those who responded "don't know" are assigned
the modal age of 65. A six-category classification of expected age is
Cohort. This variable is defined as three distinct groups: ages
45-49, 50-54 and 55-59 in 1966. This variable is to be used as a control
since both the expected and actual age may differ across cohorts, as may
the relationships between the two dependent variables and predictor vari-
Constraint factors. The first type of influence on the expected and
actual age of retirement to be considered concerns factors that restrict
the continued labor force participation of men. There are two constraint
variables measured in this research: coverage under a mandatory retire-
ment policy and work-related health limitation.
1. Mandatory Retirement Policy. Most of the literature re-
garding mandatory ages of retirement is polemical and suggests that mandatory
ages reduce the age at which people retire. Schulz (1976) and others
have shown that earlier-than-preferred retirement is a relatively rare
occurrence since the large majority of men covered by such policies re-
tire before the mandatory age or voluntarily retire at that age. It is,
therefore, hypothesized that men covered by a mandatory retirement policy
will expect to retire at earlier ages. However, it is not expected that
men covered by such policies will retire at earlier ages than men not
covered. Respondents in the NLS panel were asked in a number of waves
if they were covered by a mandatory retirement policy at their present
job. This variable is defined as a dichotomy, covered/not-covered.
2. Work-Related Health Limitation. The ability to work is
perhaps the most crucial constraint factor. Previous research consis-
tently shows that men in poor health retire at earlier ages. The hypoth-
esis is that men in poor health will expect to retire at earlier ages and
will retire at earlier ages than healthy men. Health of respondent can
be consistently measured over the waves only as a dichotomy of health
limitation/no health limitation. This is because only the basic ques-
tions asking men if their health limited the amount or kind of work they
could do were asked in all of the surveys.
Job-related factors. The second general type of influence on the
expected and actual age consists of factors directly related to the job
the individual holds. There are three variables contained in this type:
occupation, pension eligibility and satisfaction with job.
1. Occupation. The kind of occupation an individual is em-
ployed in may affect both the expected and and the actual age of retire-
ment. The research literature indicates that higher status occupations,
such as professionals and businessmen, both expect to retire, and do
retire, at later ages. Also, farmers tend to have higher expected and
actual ages of retirement. Occupational status, measured by the Duncan
Socio-Economic Index, should be positively related to both the expected
and the actual age of retirement. An 11 category classification of
occupations will also be used to investigate specific occupational
groups, such as farmers.
2. Pension Eligibility. Barfield and Morgan's research (1969,
1978) indicate that men who will receive a private pension (other than
social security) expect to retire earlier, and do retire earlier, than
men who are not eligible for pensions. It is hypothesized then, that men
who are eligible for a private pension will expect to retire earlier and
will have younger ages of retirement than those men who are not eligible.
In the 1966 survey, the following question was asked, "Will you be eligible
for any retirement benefits, such as from personal plans, private employee,
government employee or military retirement plans?" If the man responded
that he was eligible for one or more pension benefits, he is considered
eligible. The amount of pension benefits may vary widely but there is
not adequate information on this aspect of receipt of pensions. There-
fore, this variable is defined as a dichotomy, eligible/not-eligible.
3. Job Satisfaction. Previous research indicates that men
who are dissatisfied with their job expect to retire at younger ages com-
pared to men satisfied with their job. The evidence on how job satisfac-
tion affects the age of retirement is less clear, but if any conclusions
were to be reached, it does appear that dissatisfaction with job may lead
to earlier retirement. In this research it is expected that dissatisfac-
tion with job will result in earlier expected and actual ages of retire-
ment. In all waves except the 1973 and 1975 surveys, the employed
respondents were asked about satisfaction with present job. This vari-
able will be defined as a dichotomy, satisfied/dissatisfied.
Social and psychological factors. The third type of influence con-
cerns those social and psychological factors that are not directly related
to the job of the individual but may have some impact on the expected and
actual age of retirement. This third area contains nine different vari-
ables: marital status, number of dependents, education level, wife's
labor force status, race, assets, mortgage debt, the Rotter scale of
internal-external locus of control and commitment to work.
1. Marital Status. There is some evidence that married men
remain in the labor force somewhat longer than men who are not married.
It is believed that the additional responsibility of a spouse tends to
keep men in the labor force longer. Therefore, married men should have
later expected and actual ages of retirement. In the NLS data, there are
five marital status categories: married, widowed, divorced, separated
and never married. For most analyses, all of these categories will be
used; however, in some cases a married/not currently married dichotomy
will be used.
2. Number of Dependents. Previous research shows that having
more dependents results in later expected ages and also delays the age of
retirement. Having the financial responsibility of dependents should
delay retirement and, therefore, it is hypothesized that number of de-
pendents will be positively related to the expected and actual age of
retirement. The number of dependents is defined as those persons whom
the respondent claims as dependents in addition to his wife. This infor-
mation was obtained in every survey.
3. Education. Those persons with more years of education
often have greater attachment to their work, such as professionals. This
attachment may result in later expected and actual ages of retirement.
The education variable is defined as number of years of school completed
by the respondent.
4. Wife's Labor Force Status. There has been very little re-
search on how the employment of the wife may affect the retirement expec-
tations or retirement timing of men. The recent research by Anderson et
al. (1980) indicates that labor force participation by the wife tends to
result in men remaining in the labor force longer. Therefore, it is
hypothesized that those men with working wives will expect to retire
later and actually retire at later ages than men whose wives do not work.
A question concerning the earnings of the respondent's wife was included
in all waves except 1973. The variable will be measured as a dichotomy,
working (reported earnings)/not-working (no reported earnings).
5. Race. Net of all other factors, it is doubtful that there
will be differences between blacks and whites in the expected age of re-
tirement. However, the greater employment instability of blacks and
possible discrimination against blacks in the labor market may produce
younger ages of retirement for this group.
6. Assets. There have only been a few studies that have in-
vestigated the effect of assets on retirement and this evidence is unclear.
Higher assets may lead to younger expected ages of retirement and also to
younger actual ages of retirement, but some analyses show no effect of
assets on retirement. The hypothesized relationship is that assets is
negatively related to both the expected and actual age of retirement.
Information on asset holdings was obtained in 1966, 1969, 1971 and 1976.
Assets are defined as the sum of savings, savings bonds, estimated market
value of stocks and bonds, business assets and the estimated value of
real estate other than home. Business debts and debt on real estate is
subtracted from this sum.
7. Mortgage. As with assets, there has been relatively little
research on the effect of mortgage debt on retirement. The few studies
reported in the literature review indicate that the existence of a mort-
gage does delay the expected and actual age and, therefore, a positive
relation among these variables should exist. Information on the amount
of mortgage on a home was obtained in 1966, 1969, 1971 and 1976. Those
respondents who do not own a home are coded 0.
8. Rotter Scale. One of the specific factors that Atchley
(1979) discusses in his model of the retirement process is "orientation
to planning." This orientation may be related to an inner- versus other-
directed personality where those persons who are inner-directed tend to
plan more than those who are other-directed. Because those who are
"planners" should be less likely to respond "never retire" or "don't know,"
they would tend to have lower expected ages of retirement. Although this
is less certain, those persons more inner-directed should also have younger
actual ages of retirement. The Rotter scale is an 11-item measure of an
individual's internal-external locus of control. This scale was developed
to measure the individual's perception of the amount of control he has
over his own life. Those who are more "internally located" should be
more likely to plan for retirement. This is only an indirect measure of
"orientation to planning," but is the closest measure to this concept
available in the NLS data set. This scale was not included in the
questionnaire until 1969, and was included again in the 1971 and 1976
surveys. The 1969 scores will be used in the analysis of expected and
9. Commitment to Work. One of the social-psychological fac-
tors in Atchley's model is "strength of work ethic." This type of factor
measures the degree to which an individual believes in work for the in-
trinsic value of working. Those who have a strong work ethic should have
later expected and actual ages of retirement. In the 1966 survey, the
following question was asked: "If, by some chance, you were to get enough
money to live comfortably without working, do you think that you would
work anyway?" Those men who responded affirmatively are considered to
be committed to work and should expect to retire later and actually re-
tire at later ages than those men who would not continue working. This
variable is defined as a dichotomy, committed/not-committed.
It is believed that most of the important factors related to retire-
ment are measured in this research. One specific area, attitudes toward
retirement, cannot be sufficiently measured because no information on
these attitudes was obtained on a regular basis in the NLS data. The
expected age of retirement includes to some degree the preferred age of
retirement. However, the expected age primarily encompasses the realis-
tic plans of individuals and their consideration of restrictions and
opportunities regarding the timing of retirement. No claim is made
that the expected age of retirement is a measure of attitudes toward
Factors in the Analysis of Adjustment to Retirement
The two measures of adjustment are discussed first. Independent
variables for the full sample contain factors that measure "quality of
life," status, social characteristics and psychological orientation,
which are measured for both retirees and those men still in the labor
force. Finally, a set of variables that are specific to the conditions
of retirement are presented.
Happiness with life. The first dependent variable is a global hap-
piness with life measure. The following question was asked of all re-
spondents in 1976: "Taking things altogether, would you say you're very
happy, somewhat happy, somewhat unhappy or very unhappy these days?"
This question is taken from Gurin, Veroff and Feld's (1960) study of
mental health in America.
This life happiness measure is not the same as life satisfaction,
although the two are conceptually related and are also highly correlated.
Happiness with life tends to reflect the individual's affect (feelings)
about his current state of affairs (George and Bearon, 1980). The use
of one item to measure how happy a person is with life is questioned by
some researchers but has been used quite frequently in previous research.
Lohman (1977) reports that correlations of a single life satisfaction
measure with six scales of satisfaction among older people were relatively
low. Although the global happiness item refers to present circumstances,
Robinson and Shaver (1973) report studies that show relatively high test-
retest correlations of single-item measures with up to two years separ-
ating the measurements. This indicates that the responses are not
usually subject to ups and downs in daily life. Spreitzer and Snyder
(1974) defend the validity of a single item because it is straightfor-
ward while the use of scales often assumes that certain components of
happiness or satisfaction are equally important to the individual in
determining overall happiness with life. The straightforward question,
such as the one used in this research, allows the individual to evaluate
his life in a wholistic manner. The happiness with life measure will be
defined as a dichotomy of happy/unhappy due to the skew towards the "very
happy" and "somewhat happy" responses.
Evaluation of retirement experience. The second measure of adjust-
ment to retirement is specific to the retirement experience. In the 1976
survey, all retirees were asked, "All in all, how does your life in retire-
ment compare with what you expected it to be? Is it much better, somewhat
better, about what you expected, somewhat worse or much worse?" This
question asks the respondent for a relative evaluation of his retirement
experience. In general, such an evaluation will probably be congruent
with an evaluation of retirement that refers only to present satisfaction.
This "relative evaluation of retirement" item is considered to be a useful
measure for the analysis of panel data because it is more relevant to cer-
tain conditions of retirement, such as earlier-than-expected retirement,
the level of earnings replacement and the worsening of health. Those men
who responded "much better" and "somewhat better" are considered to have
a "positive evaluation" of retirement, those men who responded "about
what expected" have a "neutral evaluation" of retirement, and those men
responding "somewhat worse" or "much worse" have a "negative evaluation"
of retirement. These three types of responses--positive, neutral and
negative--constitute the categories of the evaluation of retirement
Independent variables for full sample.
1. Health. The general health of the older person has been
shown to be the most important predictor of satisfaction or happiness
with life. Those men in better health should be more likely to be happy
with their life and also to have a positive evaluation of retirement. A
basic categorical contrast of men who report one or more health problems
in 1976 to those who do not report any health problems is used to meas-
ure overall health. This variable does not measure the degree of ill
health, such as whether a man is disabled, but it does differentiate
between healthy men and men who have some health problems.
2. Change in Health. Another important aspect of a person's
health is the relative state of health at the present time compared to
the past. This kind of factor has rarely been tested among older people,
but it is clearly important for personal happiness. A recent worsening
of health may depress an individual, even if his overall state of health
is relatively good. Also, improvements in health by persons who are un-
healthy may result in more positive evaluations of life, and of retire-
ment. In the 1976 survey, the following question was asked of all
respondents: "During the past three years, has your health condition
become better, worse or remained about the same?" These three responses
are used to measure the relative state of health of the respondents.
3. Family Income. Previous research shows that higher income
results in higher levels of satisfaction or happiness with life. There
should also be a positive relation between income and the evaluation of
retirement because higher income allows for a better standard of living
and the opportunity to engage in certain activities, such as travel, in
retirement. The measure of family income is for the year 1975 and in-
cludes all sources of income for members of the respondent's household.
For those persons who are missing on this variable, a regression equation
is used to predict family income.
4. Occupational Status. The research literature indicates
that occupational status or prestige is positively related to happiness
with life and this should also be true for the evaluation of retirement.
Upper status workers tend to have greater autonomy, tend to receive more
nonpecuniary rewards and to live and work in more appealing environments.
For those who are retired, higher occupational status reflects more con-
tacts in the community, a greater retention of work contacts, more partic-
ipation in voluntary organizations and a greater variety of leisure time
pursuits. The Duncan SEI measure of occupational status is used to meas-
ure this factor.
5. Education. Most of the previous research indicates that
education level, net of occupation, is not a significant predictor of
happiness with life. There is some evidence, however, that education is
positively related to happiness with life and it is hypothesized in this
research that higher education will result in a greater probability of
happiness with life and a greater probability of a positive evaluation
6. Labor Force Status. One of the main issues in the retire-
ment literature is whether the loss of job results in lowered satisfaction
and happiness with life. As noted in the literature review, most empirical
research does not support this hypothesis. The possible negative effect
of being retired on happiness with life will be tested in this research.
The variable is a dichotomous contrast, retired/not-retired.
7. Marital Status. There is some evidence that married men
are more happy with their lives than nonmarried men. Very often men who
are divorced, separated, widowed and never married are categorized together
into a nonmarried classification. Therefore, the cause of this difference
in the tendency to be happy is not clearly understood. It would be ex-
pected that for men, as for women, widowhood is the most negative of non-
married statuses. Furthermore, the recency of widowhood may also be
important in the adjustment of the individual. In the 1976 survey,
widowed men were asked the year in which they did become widowed and
this information will be used to separate long-term widowers (more than
two years) from recent widowers (two years or less). Therefore, four
categories of nonmarried men--divorced or separated, never married, long-
term widowers and recent widowers--will be contrasted with married men.
All four nonmarried groups should have lower probabilities of being
happy, with the recent widowers being most likely to be unhappy. The
effect of marital status on the evaluation of retirement will also be
tested, but this effect should be smaller and perhaps nonexistent.
8. Rotter Scale. The description of this psychological meas-
ure of internal-external locus of control was given in the previous sub-
section on social and psychological variables related to the expected
and actual age. Those men more inner-directed are expected to have a
higher probability of being happy with their life and also more likely
to have a positive evaluation of retirement.
9. Race. Most of the research on satisfaction or happiness
with life does not show differences between blacks and whites, net of
other factors. It is expected that blacks are more likely to be unhappy
with their life and to have a negative evaluation of retirement. This
is expected because of the generally more depressed living environment
for blacks and the restricted opportunities for social participation.
1. Difference in Actual and Expected Age. The primary motiva-
tion for the analysis of the adjustment to retirement measures is to test
the effect of the discrepancy between the expected and actual age of re-
tirement. It has been hypothesized that those men who retire earlier
than they expected will have a "less successful" adjustment to retire-
ment than on-time retirees. Those men who retire later than their ex-
pected age may also have a less successful adjustment to retirement. In
terms of the specific dependent measures, those who retire "on-time"
should have a higher probability of being happy and also a greater proba-
bility of a positive evaluation of retirement. Three categories are
constructed to measure the difference in actual and expected age.
Earlier-than-expected retirees are those who retired more than two years
before their expected age; on-time retirees are those who retired within
two years, plus or minus, of their expected age; and later-than-expected
retirees are those who retired more than two years after their expected
2. Years Retired. Some previous research indicates that ad-
justment to retirement may be less successful shortly after the time of
retirement. This adjustment period has been suggested to last about one
year (Streib and Schneider, 1971). It is hypothesized that number of
years retired will be positively related to both probability of being
happy and also a positive evaluation of retirement. This variable is
measured two ways: the interval-level measure of number of years since
retirement and the categorical contrast of retired one year or less to
retired more than one year.
3. Commitment to Work. Those men who are committed to work,
whether their retirement was voluntary or not, may have a less success-
ful adjustment to retirement. Therefore, it is hypothesized that those
men who were committed to work before retirement will have a lower proba-
bility of being happy with life and more likely to have a negative evalu-
ation of retirement. This variable is a dichotomy as defined in the
previous subsection on the expected and actual age of retirement.
4. Income-Replacement Ratio. There have been very few studies
that have data on pre-retirement income because panel data is necessary
for the accurate collection of this information. Results from the Cornell
panel study (Streib and Schneider, 1971) did indicate that the drop in
income after retirement tended to reduce the life satisfaction of re-
tirees. It is expected that the higher the replacement ratio, the higher
the probability of being happy with life and of evaluating the retirement
experience positively. This measure is calculated by using the family
income measure for 1975 as the post-retirement measure of income and
taking the average family income of retirees before the year of their
retirement. For those men who retired in 1967, the family income in 1965
and 1966 are averaged; while for later retirees, additional years (1968,
1970 and 1972) are also averaged. Family income before 1975 is normed
to 1975 dollars with the use of the Current Price Index (U.S. Bureau of
the Census, 1980) in order to control for changes in money income.
Methods of AnaTysis
In this section, the statistical methods of analysis are delineated.
A number of parametric statistical methods will be used in this research,
with most models being tested with OLS regression.
One of the theoretical questions in this research concerns the cor-
relation between a behavioral expectation (expected age of retirement)
and behavior (age of retirement). Bivariate regression is used to test
this relationship, because it allows for the introduction of the bias-
correction term (x). The regressions will be conducted with both the
interval and the categorical definitions of expected age. The regression
of actual age on the categories of expected age will allow for the test-
ing of nonlinearity in the effect of expected age on actual age and also
give some indication of the viability of the assigned ages for the "don't
know" and "never retire" groups.
Also of interest is the stability of expected age of retirement over
time. The stability of expected age primarily involves an analysis of
the association between 1966 response and 1976 response. In the bivari-
ate regression analysis, the assignment rules for the "don't know" and
"never retire" groups, which were explained earlier, are used. To further
investigate consistency over this ten-year period, a log-linear analysis
is also conducted. The six-category classification previously delineated
will be used for the analysis of expected age in 1966 and expected age
in 1976. Log-linear analysis allows one to test models of association
between two or more variables. An attempt is made to fit the observed
cell frequencies (fij) with the most parsimonious and theoretically
meaningful model. With two variables (expected age at time 1 and at
time 2) there are only two models that can be tested, short of a satur-
ated model where all cells are fit exactly. The model of independence
posits no relation between the two variables and may be expressed
In(mij) = u + x + X, (1)
where ln(mij) represents the natural log of the sample expected fre-
quency for the cell ij, u is the grand mean of the log m.., x is the
effect of being in the ith row and X. is the effect of being in the jth
column. The row and column effects are deviations of each row and column
total from the grand mean (u). No relationship between the row variable
(expected age in 1966) and the column variable (expected age in 1976) is
allowed. This model is not expected to fit the observed data.
The quasi-independence model is one that allows for association in
some particular cells but maintains independence between the row and
column variables in the remaining cells. This is accomplished by requir-
ing the "dependent cells" to be fit exactly (m.. = f..) and using an
iterative procedure to calculate expected frequencies for the independent
cells. The quasi-independence model is used to postulate dependence
along the main diagonal (i = j) and independence in the off-diagonal
cells (i f j). This model may fit the observed data if there is no
definite pattern of change in expected age (earlier or later ages) over
time. If this model does not fit, the standardized residuals can be
computed, i.e., (fij-mij)/mij to investigate where association among
the categories still exists.
OLS Analysis of Expected and Actual Retirement Age
OLS estimation techniques are used to test models of expected age
and actual age of retirement. The effects of the independent variables
on the expected and actual age are analyzed in terms of their tendency
to produce larger or smaller differences between the actual and expected
The analysis begins with a prediction equation for the expected age
of retirement among the subsample of retired men. The full prediction
equation for the expected age is
ExpAge = a + bi MandRet + b2 HlthLimit + b3 Occup +
b4 PenEligib + bs JobSat + b6 Marital + b7 No.Dep's +
b8 Educ + b9 Wife'sLFS + blo Assets + b11 Mortgage +
b12 Rotter + bl3 WorkCmmt + b14 Race + Lambda. (2)
Interactions between race and other predictors, and between the mandatory
retirement variable and other predictors, will be estimated for the ex-
pected age. The analysis of the expected age of men who have retired
allows for the estimation of the effects of characteristics that are also
used in the analysis of the actual age. The effects of some variables
on the expected age may be different from their effects on the actual
The prediction of the actual age of retirement will be used in con-
junction with the prediction of the expected age to ascertain the factors
that increase or decrease the difference between the expected age and
the actual age of retirement. The regression of actual age on expected
age produces a residual that is explained in part by the previous predic-
tors of expected age, net of the intercorrelation between expected age
and the other predictors. In this sense a causal path model is estimated
in which the effects of social and economic characteristics at an earlier
time on the age of retirement are measured through their effect on the
expected age of retirement, and their effect net of the expected age.
The prediction equation for the actual age of retirement is
Actual = a + bl ExpAge + b2 MandRet + b3 HlthLimit
+ b4 Occup + bs PenEligib + b6 JobSat + b7 Marital +
b8 No.Dep's + b9 Educ + blo Wife'sLFS + bll Assets +
bl2 Mortgage + b13 Rotter + b14 WorkCmmt + b15 Race
+ Lambda. (3)
The above prediction equation proposes that characteristics and
expectations and attitudes at one point in time influence behavior at a
later period. One of the most important questions in this research, both
theoretically and practically, is how changes in individual characteris-
tics or situations may produce differences between expectations and be-
havior. These intervening changes may in fact be more important than
the characteristics or situations of individuals at a previous time.
Therefore, a complete model must include change terms for some of the
predictor variables. The prediction equation for the full model will
be in the following form:
Actual = a + b, ExpAge + b2 .. + b15 Race +
b16A HlthLimit + b17A Occup + b18A Marital + b19A
No.Dep's + b20A Assets + b21A Mortgage + b22A
Wife'sLFS + Lambda. (4)
The x regressor term must be estimated for the subsamples of retired
and not-retired men. Logit estimation allows one to calculate the prob-
ability of sample inclusion based on a set of exogenous variables. The
logit equation for the prediction of retirement in 1976 is
log( Pr ) = a + bi Age + b2 MandRet + b3 HlthLimit
+ b4 Occup + b5 PenEligib + b6 JobSat + b7 Marital +
b8 No.Dep's + b9 Educ + blo Wife'sLFS + b1, Assets +
b12 Mortgage + b13 Rotter + b14 WorkCmmt + b15 Race +
b 6 ExpAge. (5)
The probability of remaining in the labor force will be calculated
in the same manner, with the same exogenous variables, but where "work-
ing" = 1.
Analysis of Adjustment Measures
Logit estimation will be used in the analysis of the happiness meas-
ure (happy-unhappy) that was previously discussed. The happiness with
life measure is defined as a dichotomy between "happy" and "unhappy."
The logit prediction equation for the full sample (retired and not-
retired), where "happy" = 1, is
log(T r-) = a + bI HlthProb + b2 HlthBttr + b3 HlthWrse
+ b4 FamInc + b5 Occup + b6 Educ + b7 LFS + b8 DivSep +
b9 LongWid + blo RecentWid + b11 Never + b12 Rotter +
b13 Race, (6)
where HlthBttr and HlthWrse represent the effects of change in health
versus no change, LFS is the dichotomy of retired/not-retired and the
four marital statuses are represented in contrast to married men. An
interaction model for retired and not-retired men will also be tested.
Among men who are retired, the analysis of the satisfaction measure
will include more exogenous variables that are specific to being retired.
The income replacement ratio variable, the categorical contrasts of the
discrepancy between the actual and expected age, the number of years in
retirement and commitment to work before retirement are the additional
variables. These variables will be added on to the model for the full-
sample given in equation (6).
The analysis of the subjective evaluation of the retirement experi-
ence is difficult because there are three categories, and though they are
ordered in their degree of "positiveness," interval-level measurement
cannot be assumed. The difference between a "worse than expected" re-
tirement experience and an "about what expected" retirement experience
is not necessarily the same difference as that between a "better than
expected" and an "about what expected" retirement experience. The way
such responses are analyzed is by logit contrasts. In these logits
the denominator (1-Pr) is always the neutral response of "about what
expected." The positive and negative responses are, in turn, contrasted
to the neutral response. Therefore, logit equations will be estimated
for the contrasts PI/P2 and P3/P2, where P1 = "positive evaluation,"
P2 = "neutral evaluation" and P3 = "negative evaluation." The same
predictors are used in the analysis of the evaluation of retirement
measure as for the analysis of the happiness measure.
For the analysis of the happiness with life and retirement experi-
ence variables in the subsample of retired men, it will be necessary to
control for possible sample bias with the use of a X regressor term.
In this case, it is necessary to estimate the probability of being
retired with the basic set of predictors used in the full-sample analy-
sis of the happiness measure. Family income is not included, however,
because it is partially a function of retirement. Adding the necessary
factor of age to the exogenous variables, the logit estimation equation
for the prediction of retirement in 1976 is,
log(lPr) = a + bl Age + b2 HlthProb + b3 HlthBttr
+ b4 HlthWrse + b5 Occup + b6 Educ + b7 DivSep +
b8 LongWid + bg RecentWid + blo Never + b11 Rotter
+ b12 Race.
The x term will then be estimated from the predicted probabilities of
inclusion and entered into the logit equations for the analysis of the
happiness and evaluation of retirement variables.
In this chapter, the hypothesized relationships among the variables
and how these variables are to be measured have been delineated. The
methodological difficulties in the use of a panel survey have been dis-
cussed and the resolutions of the major problems have been presented.
The statistical methods of analysis have also been outlined and show the
way in which the dynamic process of the movement from a certain expecta-
tion regarding the timing of retirement to the actual timing of retire-
ment, and the possible attitudinal response to this experience will be
evaluated. Although not all of the possibly relevant factors can be
taken into account in the foregoing analysis, it is felt that the models
proposed in this chapter address the central sociological issues of the
ANALYSIS OF EXPECTED AND ACTUAL AGE OF RETIREMENT
The main issue considered in this research is the discrepancy between
the expected or planned age of retirement and the actual age of retire-
ment. As stated in Chapter I, this discrepancy is important for a number
of reasons. First, the strength of the association between the expected
age and the actual age of retirement gives some indication of how well
men are able to predict the age at which they retire. This may be con-
sidered an indirect indicator of the extent older men make plans for re-
tirement and are able to carry out such long-range plans. In a theoreti-
cal context, the relationship between expected age and actual age tests
the usefulness of behavioral intentions as predictors of actual behavior
over long periods of time. Another important reason for considering the
discrepancy between expected and actual age is the possible effect that
such a discrepancy might have on personal adjustment to retirement. This
issue will be addressed in the next chapter.
While some association is expected between the expected and actual
age, a relatively large discrepancy between the two may occur for many
men. The reasons for such discrepancies are the main focus of the fol-
lowing analysis. Frequent change in expected age may occur between first
measurement of the expected age and the time of retirement. The insta-
bility in expected age may be caused by those factors that have been
hypothesized to affect the relationship between the expected age and
the actual age. Instability in expected age may also reflect the uncer-
tainty and tentativeness of the responses of men in late-middle age.
That is, changes in expected age over time and much of the discrepancy
between expected and actual age may reflect a widespread lack of serious
planning for the timing of retirement among men approaching retirement
age. The analysis of the discrepancy between expected and actual age
centers on the issue of what factors and events influence whether men
retire at the age at which they expect to retire. Health is expected to
be a major factor in the degree of discrepancy between expected and actual
age. Of more interest from a sociological viewpoint are the effects of
social and social-psychological characteristics and the effects of changes
in these characteristics on the ability of men to retire at or near their
This chapter addresses explicitly the issues discussed above. First,
summary descriptive statistics regarding the expected and actual age of
retirement will be presented. Second, the relation between expected and
actual age will be investigated through bivariate regressions. Third,
the stability of expected age over time will be analyzed by using the
responses of men who have not yet retired. Fourth, the expected age of
the retired men will be analyzed in order to discern the types of factors
most important in determining such a behavioral intention. Finally, the
discrepancy between the expected and actual age will be analyzed.
Prediction of Sample-Bias Term
The problem of sample bias was discussed in the preceding chapter
and a method was presented in which a "bias-correction" variable is esti-
mated (lambda). Sample bias may affect the observed relationship between
the expected age and the actual age of retirement as well as the relation-
ship between these two variables and the predictor variables. The first
step in the estimation of the bias-correction term is to obtain predicted
probabilities of retirement, or of remaining in the labor force, for
each individual. The LOGIST procedure in the SAS computer package is
used for this, as well as all other logit analyses. This procedure uses
a maximum likelihood procedure to fit the logistic multiple regression
model. The logit regression for the prediction of being retired in 1976
is presented in Table 4-1. A total of 16 exogenous variables that are
used in the following analysis are included in the logit equation. Nine
of the sixteen exogenous variables significantly reduce the chi-square
value for the model. The predicted proportional change presented in the
table is the percent increase or decrease in the probability of retire-
ment for a one-unit change in the exogenous variable when evaluated at
the predicted mean of the dependent variable. As with OLS coefficients,
the proportional change should be interpreted with caution because the
units of measurement for the exogenous variables are quite different.
Age, occupation, number of dependents, assets, mortgage, education,
Rotter scale and the expected age of retirement are all interval leval;
the remaining variables are dichotomous contrasts. Age produces the
largest reduction in the likelihood-ratio test statistic, which indicates
that it has the greatest effect on the probability of retirement. The
Logit Regression for
Retirement, n = 3348
Variable Logit S.E. Change in Yt
Age .288* .013 .060
Hlth Limit. 1.662* .094 .349
Occupation -.001 .002 -.0002
Job (Dis)Satisf. -.095 .176 -.020
Work Cmmt. -.378* .103 -.079
Marital Status .140 .126 .029
No. of Dep's. -.083 .047 -.017
Assets ($1,000) -.001 .001 -.0002
Mortg. ($1,000) -.007 .007 -.001
Mand. Retire. .284** .113 .060
Pen. Eligib. .576* .111 .121
Wife Works -.373* .106 -.078
Race .274** .112 .057
Expected Age -.047* .010 -.010
Education -.047* .015 -.010
Rotter Scale .005 .009 .001
* Significant at .01 level.
**Significant at .05 level.
The predicted proportional change is the instantaneous rate of change
in the dependent variable for a one-unit change in the exogenous vari-
able. The proportional change varies depending upon where the dependent
variable is evaluated along the logistic distribution. For this equa-
tion the predicted proportional change is evaluated at the predicted
mean of the dependent variable, P = .301. The predicted mean is cal-
culated by the formula P = a + (bX71) + (b2Y2) + + j(bXY). The
equation for the predicted proportional change is b [P(1-P)T, where
bk is the logit coefficient for the kth exogenous variable.
existence of a work-related health limitation is also important in the
prediction of retirement.
The logit regression for the prediction of being in the labor force
in 1976 (not retired) produces the same logit coefficients as the pre-
diction of being retired but with opposite signs. The resulting lambda
estimates from the predicted probabilities for the two dependent vari-
ables are not perfectly negatively correlated however (r = -.927), since
lambda is not a linear function of the probabilities, but is in fact
curvilinear. The lambda estimate from the prediction of retirement is
used in the equations that involve the subsample of men who are retired.
In the analysis of the stability of expected age over time, the lambda
estimate obtained from the prediction of remaining in the labor force
Descriptive Statistics of Expected and Actual Age
The mean, standard deviation and range of the expected age and the
actual age of retirement as well as percentage distributions of expected
age are presented in Table 4-2. For the expected age the mean, standard
deviation and range are presented when the "don't know" and "never re-
tire" respondents are excluded. The "full group" includes the "don't
know" and "never retire" respondents. The "don't know" group is assigned
age 65. Members of the "never retire" group are assigned their life
expectancy based on age and race (Public Health Service, 1975). The
average age of retirement is much lower than the average expected age
in 1966 for those already retired; the mean difference is -4.66. When
the sample is separated by cohort, the average age of retirement varies
widely. Men aged 45-49 in 1966 have an average age of retirement of
Descriptive Statistics of Expected and Actual
Retirement Age for Retired and Not-Retired Men
Subsample Mean Std. Min. Max.
Age of retirement (n=1246)
Expected age (1966)
full group (n=1246)
Exclude D.K.'s (n=1059)
Exclude N.R.'s (n=1056)
Exclude D.K.'s and N.R.'s
Expected age (1966) full
Exclude D.K.'s (n=1720)
Exclude N.R.'s (n=1732)
Exclude D.K.'s and N.R.'s
Expected age (1971)
full group (n=2057)
Exclude D.K.'s (n=1794)
Exclude N.R.'s (n=1673)
Exclude D.K.'s and N.R.'s
Expected age (1976)
full group (n=2084)
Exclude D.K.'s (n=1640)
Exclude N.R.'s (n=1688)
Exclude D.K.'s and N.R.'s
of Categories of Expected Aqe
54.9; the middle cohort exhibit an average age of 60.1, while the oldest
cohort of men aged 55-59 have an average retirement age of 63.5 years.
The three cohorts reflect three different types of retirees. The young-
est cohort represents very early retirees; the middle cohort usually
contains early retirees; and the oldest cohort has an average retirement
age that is closer to "normal" retirement ages. It is apparent that in
analyzing the discrepancy between the expected and actual age the effects
of the predictors will be interpreted as either increasing the negative
discrepancy between actual and expected age or reducing this negative
The average expected age in 1966 of those who have retired and the
average expected age in 1966 of those who have not retired are only
slightly different. In the not-retired subsample there are only small
differences in the mean expected age over the years 1966, 1971 and 1976.
This indicates that there is not a strong tendency for the expected age
to increase as the cohort ages. In order to understand why the rise is
so small over the ten years, the expected age is grouped into six cate-
gories and the percentage distribution is compared across years.
At the bottom of Table 4-2, the percentage distribution of expected
age indicates that men who have retired had similar expectations in 1966
to those who did not retire. There is a larger proportion of retired men
who expected to retire between 62 and 64, whereas a slightly larger pro-
portion of men who have not retired respond that they never expect to
retire or don't know. Among those men who have not retired by 1976,
there are some definite shifts in the distribution of expected age over
time. The aging of the sample results in a lower proportion of respon-
dents expecting to retire before 62 in 1976 than in previous years, but
a much larger proportion expect to retire between 62 and 64. Somewhat
surprisingly there is also a large decrease in the proportion of men who
expect to retire at 65 in the 1976 survey. This trend away from the
"normal age" of 65 may indicate that those who are closer to the retire-
ment decision specify ages that are more realistic for them rather than
the "normal age." Movement away from age 65 may also indicate a waning
of the normal age itself and the tendency for younger or older ages to
be more acceptable for the individual and for society. However, the
larger proportion of "don't knows"in 1976 may indicate increasing un-
certainty with older age.
The Relationship Between Expected and Actual Age
The attitude-behavior approach to the analysis of individual behav-
ior predicts a relationship between intended behavior (expected age) and
behavior (actual age). The psychologists Fishbein and Ajzen (1975)
expect this relation to be a close one, because there is a high degree
of congruency in the operationalization of the two measures. Sociologists
such as Acock and DeFleur (1972) propose that such a relationship is
highly contingent on intervening situational or social factors. The
simple bivariate relationship between actual and expected age is used
as a descriptive indicator of the ability of individuals to predict
their age of retirement. The presence of a strong relation or the lack
of one may both be due to characteristics of the individual and changes
over time that cannot be planned. The analysis of these other inter-
vening factors will be considered later.
In Table 4-3, the OLS coefficients and the standardized coeffi-
cients are presented for the full sample of retired men and for various
subgroups of this sample. When all cohorts are grouped significant re-
lationships are found but the correlations are relatively small and
indicate a large discrepancy between expected and actual age. The
lambda term does alter the size of these coefficients slightly. The
age range of this sample is not only truncated to 15 years but also
results in a "young" sample of retirees. Age of retirement is highly
correlated with cohort (r = .76), and more importantly, the expected
age is correlated with cohort (r = .18). Because of the intercorrela-
tions of age, expected age and actual age, the analysis of the relation-
ship between expected and actual age by cohort results in nonsignificant
relationships within cohorts. The only exception to this is the oldest
cohort where the regression estimates become significant when those who
responded "never retire" are excluded. The strength of this relation-
ship is, however, relatively low. The general increase in correlation
between expected and actual age when the "don't knows" and "never retires"
are excluded indicates that the assigned ages for these two groups are
more discrepant from the actual age than the estimates of those individ-
uals who did specify an age.
To investigate why the assigned ages of the "don't know" and "never
retire" groups tend to reduce the positive relation between expected age
and actual age, the six-category classification of expected age is used
to predict the actual age. The results in Table 4-4 indicate that men
who expected to retire before 62 and between 62 and 64 do retire at
significantly younger ages than men who expect to retire at 65. It is
important to note that the mean ages of retirement are below the expected
Regression of Actual Age on Expected Age
for Full Group and by Age Cohort
Adjusted Adjusted Adjusted
Cohort b b w/x r r w/x R2 w/X
Full group .114* .133* .153 .178 .032
(n=1246) (.021) (.018)
Exclude D.K.'s .112* .131* .168 .196 .038
(n=1059) (.030) (.018)
Exclude N.R.'s .319* .289* .241 .219 .048
(n=1056) (.039) (.034)
Exclude D.K.'s & .341* .305* .288 .257 .066
N.R.'s (n=869) (.038) (.036)
Full group .039 .056 .083 .116 .013
(n=182) (.035) (.036)
Exclude D.K.'s .048 .063 .113 .144 .021
(n=150) (.035) (.036)
Exclude N.R.'s .017 .032 .029 .052 .003
(n=159) (.048) (.049)
Exclude D.K.'s .053 .062 .095 .109 .012
N.R.'s (n=127) (.050) (.050)
Full group -.001 -.012 -.001 -.021 .000
(n=452) (.026) (.026)
Exclude D.K.'s -.002 -.013 -.003 -.026 .001
(n=390) (.025) (.026)
Exclude N.R.'s .069 .055 .077 .060 .004
(n=389) (.045) (.045)
Exclude D.K.'s & .086 .072 .107 .088 .008
N.R.'s (n=327) (.044) (.045)
Full group .007 -.003 .016 -.008 .000
(n=612) (.018) (.018)
Exclude D.K.'s .008 -.004 .020 -.010 .000
(n=519) (.018) (.018)
Exclude N.R.'s .240* .228* .236 .223 .050
(n=508) (.044) (.044)
Exclude D.K.'s & .245* .231* .265 .247 .061
N.R.'s (n=415) (.044) (.044)
* Significant at .01 level.
Regression of Actual Age on Categories of Expected Age
Expected Age Mean Mean w/X b b w/A
<62 58.92 61.36 -2.76* -2.36*
62-64 61.05 62.96 .63 .76*
65t 61.68 63.72
>65 61.69 64.20 .01 .48
N.R. 61.04 63.87 .64** .15
D.K. 60.62 63.05 -1.06* .67**
Adjusted R2 w/X .043
* Significant at .01 level.
** Significant at .05 level.
t The "65" group is the contrast category and, therefore,
represents the intercept.
age for all groups. Those who responded "don't know" in 1966 retire at
significantly younger ages than the "65" group, which explains why the
assigned age of 65 for this group tends to reduce the correlation between
expected and actual age. The greatest deviation from a linear relation-
ship is the group responding "never retire." Although this group could
not have retired at their assigned ages (72 to 78), it was possible that
they would retire at later ages than other respondents, but this does not
occur. For the OLS equations that analyze effects of other factors on
the relationship between expected and actual age, one or both of the
"don't know" and "never retire" groups will be excluded from the analysis.
The Stability of Expected Age Over Time
One of the reasons that there is a small or no relation between the
expected and actual age of retirement may be that individuals frequently
change their expected age. This might be in response to situational
changes, additional or more accurate information about retirement, or
some individual re-evaluation process that occurs with the passing of
time. This instability may also be due to the lack of planning, includ-
ing the absence of any clear idea of the probable age of retirement,
among many of the men in the sample. Bivariate regressions are used
to estimate the relationship between expected age at one time and ex-
pected age at a later time. The years 1966, 1971 and 1976 are used to
evaluate the stability of expected age. The number of persons with an
expected age is lower in 1976 (n = 2084) and 1971 (n = 2057) than in
1966 (n = 2102). The nonresponses in 1976 are apparently due to some
not-retired respondents who were not working at the time of the survey,
being "skipped over" on the expected age of retirement question. The non-
responses in 1971 occur for two reasons. First, some respondents were
retired in 1971 but later returned to the labor force. Second, some re-
spondents were not interviewed in 1971 but were interviewed in the later
waves. These differences in sample size are not large and should not
appreciably affect the size of the coefficients.
Table 4-5 presents the results of the bivariate regressions across
three time periods. The strength of the relationship over ten years is
shown to be relatively low. This correlation increases to a moderate
level (between .40 and .60) when the time interval is reduced to five
years. The exclusion of those responding "don't know" in either of the
years increases the correlation between expected and actual age, indi-
cating that those who are undecided at one time but specify an age at
another time usually do not respond "65." When those who respond "never
retire" are excluded it is found that the correlations remain about the
same over the ten year period and decrease in the five year intervals.
As with the analysis of the relation between expected and actual
age, the subsample of men who have not retired is split into three
cohorts. OLS estimates and Pearson correlations for expected age over
three time periods within the cohorts are shown in Table 4-6. In
general, the stability of expected age is greater in the younger co-
hort and least in the oldest cohort when the "don't know" and "never
retire" groups are included. This pattern is virtually reversed when
the "don't know" and "never retire" groups are excluded. The oldest
cohort appears to have the lowest correlation in the 1971-1976 period
which probably reflects the larger effect that aging has on this cohort
compared to the younger cohorts.
Regression of Expected Age-1976 on Expected Age-1966,
Expected Age-1976 on Expected Age-1971 and Expected Age-1971
on Expected Age-1966 for Full Group and Selected Subsamples
Adjusted Adjusted Adjusted
Cohort b b w/X r r w/X R2 W/X
Expected Age-1976 on Expected Age-1966
Full group .271* .277* .272 .278 .077
(n=2084) (.021) (.020)
Exclude all D.K.'s .375* .381* .366 .372 .138
(n=1378) (.026) (.024)
Exclude all N.R.'s .257* .258* .276 .278 .077
(n=1455) (.023) (.023)
Exclude all D.K.'s & .305* .305* .322 .322 .104
N.R.'s (n=939) (.029) (.029)
Expected Age-1976 on Expected Age-1971
Full group .404* .378* .419 .388 .151
(n=2042) (.019) (.019)
Exclude all D.K.'s .558* .524* .541 .502 .252
(n=1455) (.023) (.022)
Exclude all N.R.'s .360* .346* .384 .368 .135
(n=1444) (.023) (.023)
Exclude all D.K.'s & .468* .452* .458 .441 .195
N.R.'s (n=1013) (.028) (.028)
Expected Age-1971 on Expected Age-1966
Full group .506* .510* .492 .496 .246
(n=2057) (.020) (.020)
Exclude all D.K.'s .584* .588* .584 .588 .346
(n=1535) (.021) (.020)
Exclude all N.R.'s .452* .452* .462 .462 .213
(n=1515) (.022) (.022)
Exclude all D.K.'s & .495* .494* .514 .513 .263
N.R.'s (n=1123) (.025) (.024)
* Significant at the
Regression of Expected Age-1976 on Expected Age-1966,
Expected Age-1976 on Expected Age-1971 and Expected Age-1971
on Expected Age-1966 by Age Cohort
Adjusted Adjusted Adjusted
Age Cohort b b w/x r r w/X R2 w/X
Expected Age-1976 on Expected Age-1966
Exclude D.K.'s &
Exclude D.K.'s &
Exclude D.K.'s &
Expected Age-1976 on Expected Aqe-1971
Exclude D.K.'s &
Exclude D.K.'s &
Exclude D.K.'s &
TABLE 4-6 continued
on Expected Age-1966
Exclude D.K.'s &
Exclude D.K.'s &
Exclude D.K.'s &
* Significant at .01 level.
A log-linear analysis is conducted to analyze which specific cate-
gories of expected age tend to be more stable over time and where move-
ment between categories is most common. Because of the importance of
cohort, shown through the regressions in Table 4-6, cross-tabulations
of expected age-1966 by expected age-1976 are created for each cohort.
Table 4-7 presents the standardized residuals of the independence
and quasi-independence models for the three cohorts. The independence
model assumes no relation between the two variables and the quasi-
independence model allows for association along the main diagonal (i = j),
but assumes no association in the off-diagonal cells (i # j). For all
three cohorts the independence model is rejected, indicating some asso-
ciation between the categories of expected age over time. For the 45-49
cohort and 50-54 cohort, the quasi-independence model is also rejected,
indicating some association in the off-diagonal cells. No quasi-
independence model was tested for the oldest cohort, which is necessarily
a six by four table due to the aging of the cohort.
In the youngest cohort, the stability of expected age is greatest
at the two extremes of the categorical distribution; the "before 62" and
"never retire" groups. The standardized residuals for the quasi-
independence model are in parentheses under the residuals for the inde-
pendence model. Using the standardized residuals of the quasi-independence
model, it is clear that movement tends to be to adjacent categories and
is less common across two or more categories. In the 50-54 cohort there
is less association along the main diagonal than in the 45-49 cohort,
especially in the two youngest age categories; a finding that is due to
the aging of the cohort. In this middle cohort, the off-diagonal associ-
ation tends to be stronger in the adjacent categories as was found for