This article was downloaded by: [York University Libraries]

On: 23 November 2014, At: 11:14

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number:
1072954 Registered office: Mortimer House, 37-41 Mortimer Street,
London W1T 3JH, UK

Basic and Applied Social

Psychology
Publication details, including instructions
for authors and subscription information:
http://www.tandfonline.com/loi/hbas20

Measuring Sexual
Harassment: Theoretical
and Psychometric
Advances
Louise F. Fitzgerald , Michele J. Gelfand &
Fritz Drasgow
Published online: 07 Jun 2010.

To cite this article: Louise F. Fitzgerald , Michele J. Gelfand & Fritz

Drasgow (1995) Measuring Sexual Harassment: Theoretical and Psychometric
Advances, Basic and Applied Social Psychology, 17:4, 425-445, DOI: 10.1207/
s15324834basp1704_2

To link to this article: http://dx.doi.org/10.1207/s15324834basp1704_2

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of

all the information (the “Content”) contained in the publications
on our platform. However, Taylor & Francis, our agents, and our
licensors make no representations or warranties whatsoever as to the
accuracy, completeness, or suitability for any purpose of the Content.
Any opinions and views expressed in this publication are the opinions
and views of the authors, and are not the views of or endorsed by
Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of
information. Taylor and Francis shall not be liable for any losses,
actions, claims, proceedings, demands, costs, expenses, damages,
and other liabilities whatsoever or howsoever caused arising directly
 or indirectly in connection with, in relation to or arising out of the use
of the Content.

This article may be used for research, teaching, and private study
purposes. Any substantial or systematic reproduction, redistribution,
reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access
and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [York University Libraries] at 11:14 23 November 2014
 BASIC AND APPLIED SOCIAL PSYCHOLOGY, 1995, I7(4), 425-445
Copyright 0 1995, Lawrence Erlbaum Associates, Inc.

Measuring Sexual Harassment:

Theoretical and Psychometric Advances

Louise F. Fitzgerald, Michele J. Gelfand,

Downloaded by [York University Libraries] at 11:14 23 November 2014

and Fritz Drasgow

Department of Psychology
University of Illinois at Urbana-Champaign

This article describes a program of research designed to yield a conceptually

grounded, psychometrically sound instrument for assessing the incidence and
prevalence of sexual harassment in the workplace. Following the specification
of a theoretical framework that is consistent with both legal guidelines and
psychological research, we review the development and evaluation of a
three-dimensionalmodel of sexual harassment (gender harassment, unwanted
sexual attention, and sexual coercion). Based on this model, we describe the
development of a revised version of the Sexual Experiences Questionnaire
(SEQ; Fitzgerald et al., 1988). Following extensive pilot work, the instrument
was field tested in a large regulated utility. Data from 448 employed women
(professional, technical, clerical and blue collar workers) support the reli-
ability of the scales, and confirmatory factor analysis in this new sample
confirms the stability and generalizabilityof the theoretical model. Following
a brief review of validity data recently reported in the literature, implications
for further measurement improvements are discussed.

Whatever exists, exists in quantity and can be measured. - E . L. Thorndike

It seems safe to say that no concept has come as close to defying

Thorndike's famous aphorism as sexual harassment. Although its existence
has been acknowledged in the psychological literature for well over a
decade, and interest in the topic has virtually exploded in the past few years,
the legendary passion of psychologists for measuring anything that varies
has been notable, in this instance, mainly by its absence. As a result, the
state of the art (or, in this case, the science) has remained at a fairly
rudimentary level, with the great majority of studies relying on simple

Requests for reprints should be sent to Louise F. Fitzgerald, Department of Psychology,

University of Illinois at Urbana-Champaign, 603 East Daniel Street, Champaign, IL 61820.
 426 FITZGERALD, GELFAND, DRASGOW

checklists of unknown reliability and validity, without benefit of any

theoretical framework; furthermore, such checklists generally vary from
study to study. Even very basic questions remain unanswered; for example,
what is the appropriate level of measurement? What are the dimensions that
need to be assessed? What is the appropriate relation between the legal and
psychological conceptions of harassment? These are only a few of the
questions that remain not only unanswered but, for the most part, unasked.
In this article, we describe a program of research designed to yield a
conceptually grounded, psychometrically sound instrument for assessing
Downloaded by [York University Libraries] at 11:14 23 November 2014

the incidence of sexual harassment in organizations. We begin with a brief

review of earlier work on which our present efforts are based. We then lay
out our conceptual framework, including its assumptions and the develop-
ment and testing of our theoretical model of sexual harassment. Following
a brief description of the development and piloting of our instrument, we
present findings from a large-scale organizational study and close with a
few observations about what we have learned so far, as well as the questions
that remain unanswered.

BACKGROUND AND EARLY WORK

It is perhaps an overstatement to suggest that no theoretical work has been
done in this area, as the first systematic attempt to map the conceptual
domain of sexual harassment and develop a comprehensive classification
system was initiated in 1980. Writing before any legal framework for
harassment had been articulated, Till (1980) classified the self-described
experiences of a national sample of college women into five behavioral
categories: gender harassment, seductive behavior, sexual bribery, sexual
coercion, and sexual imposition or assault. He recommended that these
categories be thought of as levels of harassment because they appeared to
form a rough continuum of severity, and he argued that they were
exhaustive as they could encompass any particular example (act) of
harassment. More recently, Gruber (1992) proposed a system composed of
11 specific types of harassment, organized into three higher order catego-
ries: verbal requests, verbal remarks, and nonverbal displays. Each of these
categories contains a subset of the 11 types of harassment (e.g., verbal
requests is composed of sexual bribery, sexual advances, relational ad-
vances, and subtle pressure/advances) which Gruber arranged in decreasing
order of severity.
In contrast to such systems, which are organized at the level of classes or
categories of behavior, the most widely used framework has been that of the
U.S. Merit System Protection Board (USMSPB, 1981, 1987), which devel-
oped a simple enumeration of seven harassing behaviors that were classified
into three levels of severity: less severe (unwelcome sexual remarks,
 suggestive looks and gestures, and deliberate touching), moderately severe
(pressure for dates, pressure for sexual favors, and unwelcome letters and
telephone calls), and most severe (actual or attempted rape or sexual
assault). Until recently, most data-collection efforts have used some version
of this checklist, asking respondents to indicate whether they have experi-
enced any of the behaviors described. Absent from this literature has been
any attempt to link such data-collection efforts to a conceptual framework
or t o ascertain the reliability or validity of the measures.
Downloaded by [York University Libraries] at 11:14 23 November 2014

Sexual Experiences Questionnaire (SEQ)

To address these issues, Fitzgerald et al. (1988) developed the Sexual

Experiences Questionnaire (SEQ), a self-report inventory representing the
first attempt to assess the prevalence of sexual harassment in a manner that
met traditional psychometric standards. Based on a content validity strat-
egy, the original SEQ was designed to tap each of Till's (1980) five
dimensions via multiple items developed through literature searches, focus
groups, and consultation with subject matter experts. All items employed a
standard stem and were written in strictly behavioral terms; the actual term
sexual harassment did not appear until the end of the inventory. This last
was considered particularly important given the ambiguity surrounding the
concept, as well as individual differences in willingness to apply the term
among women who reported similar experiences; we thus sought to avoid
confounding the experience of offensive sex-related behavior at work with
the labeling of that experience as sexual harassment. Respondents were
instructed to circle the response most closely describing their own experi-
ences on a scale with three options-never, once, and more than once (the
latter two being subsequently collapsed for purposes of scoring)- yielding
frequencies or percentages of individuals who indicated thht they have
experienced the situations included in the inventory. Sample items appear in
Table 1.
This original version of the SEQ yielded an internal consistency coeffi-
cient of .92 based on a sample of approximately 1,700 college students.
Corrected split-half reliability coefficients for the five subscales ranged
from .62 to .86 and averaged .75, whereas test-retest stability estimates
computed on a considerably smaller sample (N = 46) yielded a coefficient
of .86 over a 2-week interval. We attempted to build in content validity by
writing multiple items to assess each of the five categories, which consti-
tuted the facets of our conceptual domain. In addition, we correlated each
item with what was labeled the criterion item (i.e., the final item on the scale
that asked "Have you ever been sexually harassed?") and examined the
average item-criterion correlations for each subscale, reasoning that if Till
(1980) were correct that the dimensions formed a rough continuum of
 TABLE 1
Sample Items From the SEQ

Category Sample Items

Gender harassment Have you ever been in a situation where a supervisor or
coworker habitually told suggestive stories or offensive jokes?
Seductive behavior Have you ever been in a situation where a supervisor or
coworker attempted to establish a romantic sexual relationship
with you despite your attempts to discourage h i ?
Sexual bribery Have you ever been in a situation where you felt you were being
Downloaded by [York University Libraries] at 11:14 23 November 2014

subtly bribed with some sort of reward (e.g., preferential

treatment) to engage in sexual behavior with a coworker?
Sexual coercion Have you ever been in a situation where you actually
experienced negative consequences for refusing to engage in
sexual activity with a coworker?
Sexual imposition Have you ever been in a situation where a coworker made
unwanted attempts to stroke or fondle you (e.g., stroking
your leg or neck, touching your breast, etc.).

severity, the correlation should increase in a linear fashion across the five
subscales. With one exception, the correlations conformed to expectation,
ranging from .15 for gender harassment to .37 for sexual threats. The
coefficient for sexual imposition/assault was lower than expected, most
likely because several items showed very little variance. In general, the SEQ
appeared to possess sufficient reliability and validity for research purposes.
Over the past several years, the SEQ has been used in a number of
prevalence studies, both large and small, in a large variety of educational,
occupational, and organizational settings; it has also been translated into
other languages and used in cross-cultural settings. In this research, it has
been found to be reliable, and a small but growing number of studies have
documented its validity. Reviews and commentary have been positive (e.g.,
Arvey & Cavanaugh, 1995; Beere, 1990), and it is generally acknowledged
as the most theoretically and psychometrically sophisticated instrument
available. Research with the SEQ has also, however, demonstrated a
number of shortcomings; for example, several critical items typically
demonstrate markedly skewed base rates (specifically, the bribery and
threat items that constitute quid pro quo harassment, as well as the more
serious forms of sexual imposition), complicating the application of
standard statistical techniques and suggesting the desirability of developing
more sensitive wording. In addition, standard procedures are lacking for
computing and assigning continuous scale scores to individuals because the
original scoring methods utilized dichotomous procedures designed to
produce only frequency distributions (i.e., the percentage of individuals in
a given sample who had experienced a particular type of behavior). Finally,
some experimental work (e.g., Fitzgerald & Hesson-McInnis, 1989) sug-
 MEASURING SEXUAL HARASSMENT 429

gested the desirability of distinguishing between type and severity of

harassment, and at least one study (Fitzgerald & Shullman, 1985) cast doubt
on the five-dimensional structure, suggesting that only three dimensions
were necessary to explain variation in SEQ data. Such considerations led us
to rethink our conceptual framework and revise the instrument accordingly,
with hopes of addressing a number of these issues in the process.

SEXUAL HARASSMENT:
Downloaded by [York University Libraries] at 11:14 23 November 2014

A CONCEPTUAL FRAMEWORK

Assumptions

We began by distinguishing between sexual harassment as a legal concept

and a psychological construct, noting that the two are not completely
isomorphic. For example, women have been shown to confront a wide
range of psychologically noxious workplace experiences, most of which
constitute patterns or processes rather than events. Any particular exemplar
may embody behavioral instances of varying types, frequency, intensity,
and duration and may or may not meet current legal criteria for sexual
harassment. This is particularly so given that legal criteria evolve and
change based on regulatory definitions, case law, appellate decisions, and
the like.
For example, until the Supreme Court's decision in Meritor Savings Bank
v. Vinson (1986), it was unclear whether the concept of hostile environment
(as opposed to quidpro quo) would survive as a viable cause of action under
Title VII. Psychological constructs, on the other hand, are defined not by
fiat but rather by the relations among the variables, that is, patterned
behaviors and their nomological net. In this case, it is clear from even a
cursory examination of item correlations that both types of behavior are
dimensions of the same construct- an empirical relation that presumably
existed before the enactment of civil rights legislation and will continue to
do so independent of subsequent legislation and judicial interpretation.
This observation has a number of implications for instrument design.
First, each legal determination is always contingent on a variety of factors
not assessable by psychometric measures; it is thus important not to equate
the two. However, it is desirable from a validity perspective (and imperative
from an applied one) that the relation between them be articulated. For
practical purposes, we interpret this to mean assessing the full range of
experiences that have the potential to meet legal criteria, given an appro-
priate set of facts and circumstances, whiie at the same time acknowledging
the distinction.
Second, the notion of a construct implies that measures should contain
 430 FITZGERALD, GELFAND, DRASGOW

multiple observable indicators of each aspeqt or dimension of the construct.

This represents a marked departure from the current practice of treating
each behavior as a dichotomous variable isolated from other behaviors, and
assessed via a brief, global, and frequently ambiguous phrase (e.g., pressure
for dates)-in essence, as a separate independent construct, tapped by a
lone indicator (see, e.g., USMSPB, 1981, 1987; and many others).
Given these assumptions, our first task was to specify a theoretical model
of harassment that is consistent with both the legal framework and
psychological theory; we then subjected this model to a series of rigorous
Downloaded by [York University Libraries] at 11:14 23 November 2014

tests and, finally, developed an instrument based on this framework to

assess sexual harassment in an efficient, reliable, and valid manner. It is this
series of studies that is the subject of this article.

Theoretical Model

Over the last several years, the legal framework of sexual harassment has
become increasingly clear. The original U.S. Equal Employment Opportu-
nity Commission (EEOC, 1980) guidelines have been found by the courts to
prohibit coerced sexual exchange (quid pro quo) and other types of
generally offensive sex-related behavior that have the effect of creating a
noxious workplace environment (hostile environment). In contrast to these
broad categories, the psychological research on this topic has examined only
specific acts, with little attempt to aggregate them at any higher level of
generality. As noted elsewhere, "Virtually no theoretical attention has been
given to defining the domain of this construct, nor to specifying its structure
or dimensions . . ." (Gelfand, Fitzgerald, & Drasgow, 1995). Given these
observations, our first task was to specify the theoretical dimensions (i.e.,
the domain) of the sexual harassment construct from which suitable,
observable indicators (i.e., behaviors) could be systematically sampled. Our
conceptual model appears in Figure 1.
Specifically, we proposed that the behavioral construct of sexual harass-
ment is composed of three related, but conceptually distinct, dimensions:
sexual coercion, unwanted sexual attention, and gender harassment.
Gender harassment refers to a broad range of verbal and nonverbal
behaviors not aimed at sexual cooperation but that convey insulting,
hostile, and degrading attitudes about women. Some examples include
sexual epithets, slurs, taunts, and gestures; the display or distribution of
obscene or pornographic materials; gender-based hazing; and threatening,
intimidating, or hostile acts. The EEOC (1993) recently supplemented its
original Sex Discrimination Guidelines with a statement explicitly prohib-
iting such gender-based harassment in the workplace.' Although this

'As this article was completed, those particular guidelines (which also preclladed harassment
 MEASURING SEXUAL HARASSMENT 431
Downloaded by [York University Libraries] at 11:14 23 November 2014

Hostile Environment Quid Pro Quo

FIGURE I The relation between behavioral categories of sexual harassment and
parallel legal concepts.

appears to be the most widespread form of harassing behavior (particularly

in the nontraditional, blue-collar workplace), researchers have typically
ignored it, choosing to focus on more explicitly sexual (as opposed to
sex-related) situations.
Unwanted sexual attention is, of course, just that and includes a wide
range of verbal and nonverbal behavior that is offensive, unwanted, and
unreciprocated, whereas sexual coercion constitutes the canonical example
of sexual harassment, that is, the extortion of sexual cooperation in return
for job-related considerations. The consistency of these behavioral catego-
ries with their legal counterparts is apparent from Figure 1, which depicts
the isomorphism of sexual coercion to the legal concept of quid pro quo and
suggests that gender harassment and unwanted sexual attention constitute
the two aspects of a hostile work environment. Like all theoretical models
of human behavior, Figure 1 represents something of an oversimplification;
for example, unwanted sexual attention from a supervisor with the power to
hire, fire, and promote carries an implicitly coercive message. Similarly,
stripped of its context, it is sometimes difficult to determine whether a
sexualized conversation is a come on or a put down (the essential distinction

of other types) have been withdrawn; the EEOC is planning to reissue a separate set of
guidelines specific to gender harassment in the near future (B. Henderson, personal commu-
nication, February 5, 1995).
 432 FITZGERALD, GELFAND, DRASGOW

between unwanted sexual attention and gender harassment). Despite such

real-world ambiguities, we submit that the model captures the basic
dimensions of sexual harassment as defined by law and described by
psychological research.
In a previous paper (Gelfand et al., 1995), we described a test of this
model in three samples representing different settings (employed women
and college students) and cultures (United States and Brazil) via an
application of Joreskog's (1971) procedure for simultaneous factor analysis
in several populations. A stringent test, this procedure necessitates a priori
Downloaded by [York University Libraries] at 11:14 23 November 2014

model specification, the assignment of each item to its theoretically

appropriate factor, and the constraint that each item load in identical
fashion across all samples. Taking the SEQ as the most comprehensive
operational definition of harassment available, we examined a subset2of its
original validation sample of university women (n = 434); a second sample
(n = 389) enrolled in one of four Brazilian universities (these participants
completed a Portuguese translation); and a third sample consisting of 307
female employees including academic, professional, clerical, and blue-
collar workers. Results of the separate analyses-as well as the simulta-
neous three-sample procedure-revealed a clear and compelling three-
factor solution closely corresponding to the hypothesized model and
confirming the three-dimensional structure first suggested by Fitzgerald and
Shullman's (1985) exploratory analysis. Not only did the items load as
hypothesized in each sample, but when specifically constrained to the
idential structure across samples, the combined data continued to fit the
model closely. Based on these results, we argued that our proposed model
of sexual harassment is (a) theoretically reasonable; (b) structurally iden-
tical across both workplace and educational settings; and (c) cross-
culturally generalizable, at least across the cultures examined. A complete
account of this research can be found in Gelfand et al. (1995).

SEQ: A REVISION
Following the establishment of this conceptual framework, our next step
involved the development of a revised instrument. In addition to conceptual
and psychometric desiderata, we sought to (a) develop an instrument short
enough for practical use in organizations, (b) provide balanced item
coverage for each dimension, and (c) address base rate and associated
distributional problems through the development of more sensitive item and
scaling procedures. The initial step in this process involved the generation of
a revised item pool; in addition to the best SEQ items (e.g., those that are
'The original data set contained more than 1,700 individuals; we randomly selected 25% of
these for analysis so that this sample would not receive disproportionate weight in the analysis.
 MEASURING SEXUAL HARASSMENT 433

unidimensional, most reliable, and least skewed), new items were devel-
oped, edited, and revised. As with the original form, all items were
behaviorally based and sufficiently detailed to ensure that respondents
interpret them in a similar manner.
These procedures resulted in a total of 54 behavioral items, balanced to
the degree possible, across the three dimensions of the model. To address
distributional problems, a 5-point Likert scale was substituted for the
original 3-point scale; items were arranged so that milder forms of gender
harassment and unwanted sexual attention appeared at the beginning of the
Downloaded by [York University Libraries] at 11:14 23 November 2014

instrument, followed by sexual coercion and the more intrusive physical

forms of unwanted sexual attention. The resulting scale was then piloted as
part of a larger survey examining a variety of organizational and institu-
tional issues related to sexual harassment. The pilot sample included 150
female graduate students from the Colleges of Law, Commerce, Veterinary
Medicine and the Institute for Labor and Industrial Relations at the
University of Illinois; it is these data that were used for preliminary item
and reliability analysis. Selection criteria included basic item characteristics
such as means and standard deviations, item-subscale correlations, and
theoretical concerns (e.g., balancing verbal with nonverbal behaviors and
mild behaviors with more severe behaviors, and generally ensuring adequate
construct coverage via the fewest number of items).
These procedures resulted in a revised scale containing 20 items; in
addition to the three subscales and the criterion item, the final unwanted
sexual attention item (which meets the legal definition of attempted rape)
was retained as a separate category and not scored as part of the subscale.
Alpha reliabilities in the pilot sample were .86 for gender harassment, .75
for unwanted sexual attention, and .87 for sexual coercion; alpha
reliability for the total scale was .89. Following minor editing, this revised
scale (referred to as SEQ-W) was then examined in an extensive validation
study conducted in a large, regulated utility company.

METHOD

Sample and Procedure

The sample consisted of 1,188 employees of a west coast public utility

company who participated as part of that organization's efforts to develop
a harassment-free workplace. In addition to the SEQ-W: participants also

3A parallel form, designed for students, is designated the SEQ-E; these instruments, along
with a packet of informationdescribing their characteristicsand use, are available from Louise
F. Fitzgerald (the first author).
 434 FITZGERALD, GELFAND, DRASGOW

completed a number of other scales assessing job attitudes, job stress,

organizational tolerance for sexual harassment, organizational withdrawal,
and other attitudinal and behavioral responses to organizational and job
characteristics. An eight-person research team administered the question-
naires on-site during the course of 5 consecutive working days; data were
collected in groups ranging from 1 to 78, under conditions of guaranteed
confidentiality and anonymity. The sampling frame was the work group (as
opposed to the individual employee); work groups at various sites across the
state were randomly selected to participate.
Downloaded by [York University Libraries] at 11:14 23 November 2014

These procedures resulted in a total of 1,156 completed or partially

completed questionnaires (a 97% participation rate), of which 448 were
from female employees, including professional, technical, clerical, and
blue-coIlar workers. If a participant omitted one or two responses from the
SEQ, truncated item means were inserted in place of the missing responses.
No scale scores were computed for questionnaires that were missing more
than two responses. In addition to the SEQ-W, respondents completed a
variety of additional measures. Those results are reported elsewhere
(Gelfand & Drasgow, 1994; Schneider & Swan, 1994; Zickar, 1994);
however, the portions of that research that bear on the validity of the SEQ
are reviewed briefly in the following section.

ANALYSIS

Basic Psychometric Analyses

Alpha reliability coefficients in this sample were comparable to those

obtained in the development sample with the exception of sexual coercion,
whose value reflects the low base rate of this experience in this sample.
These correlations appear in Table 2, along with item-total, item-scale, and
scale-total correlations.

Structural Analyses

Based on both theory and our previous work, we specified a measurement

model containing five gender harassment items, seven unwanted sexual
attention items, and five gender coercion items. Table 3 displays an
abbreviated version of the items, arranged according to their theoretical
assignment. Despite improved items and scaling procedures, the persistent
base rate problems with sexual coercion necessitated that the items be
dichotomized. Thus, we employed the PRELIS (Joreskog & Sorbom, 1988)
polychoric option to compute tetrachoric correlations and their asymptotic
variance-covariance matrix; the method of weighted least squares (WLS)
was then used to estimate the free elements in the factor-loading matrix.
 MEASURING SEXUAL HARASSMENT 435

TABLE 2
Item Statistics (SEQ-W)

Item/Total r Item/Scale r
Gender harassment (a = .82)
1) ... told suggestive stories .62 .64
3) . . . made crude sexual remarks .74 .73
.
4) . . made offensive remarks .70 .61
8) .. . displayed offensive materials .51 .51
9) . .. sexist comments .59 .60
Unwanted sexual attention (a = 35)
Downloaded by [York University Libraries] at 11:14 23 November 2014

.
2) . . attempted to discuss sex .73 .59
5) .. . unwanted sexual attention .71 .72
6) ... staring, leering at you .66 .66
7) . . . attempts to establish a sexual
relationship .57 .63
10) . . . repeated requests for driiks, dinner,
despite rejection .50 .59
13) . . . touching in a way that made you feel
uncomfortable .49 .49
14) . . . attempts to stroke or fondle .39 .48
Sexual coercion (a = .42)
11) ... subtly bribed you .38 .12
..
12) . subtly threatened you .14 .10
17) . . . made it necessary to cooperate to be
well treated .27 .18
18) . . . made you afraid of poor treatment if
you didn't cooperate .25 .45
.
19) . . experienced consequences for refusing .18 .19
Note. One sexual coercion item was dropped from the analysis because of extremely
restricted variance.

Because the diagonal elements of the factor variance-covariance matrix

were not fixed, a large loading in each column of the factor matrix was fixed
at 1.O. LISREL estimates of this matrix appear in Table 4; the associated
goodness-of-fit index (GFI) measures are displayed in Table 5, which
reveals a clear and compelling three-factor solution. As can be seen, the
chi-square statistic approaches the value of its degrees of freedom, the GFI
and Adjusted Goodness of Fit Index are greater than .95, and the root mean
square residual (RMSR) is reasonably low for tetrachoric correlations, thus
providing additional support for the robustness of the theoretical model
and the construct validity of the revised instrument.

DISCUSSION

The goal of our research was to develop a conceptually grounded, psycho-

metrically sound instrument for assessing the incidence of sexual harass-
 436 FITZGERALD, GELFAND, DRASGOW

TABLE 3
Measurement Model for the SEQ-W

Unwanted
Gender Sexual Sexual
Harassment Affenfion Coercion
1) . . . told suggestive stories
3) . . . made crude sexual remarks
4) . . . made offensive remarks
8) . . . displayed offensive materials
9) . . . sexist comments
Downloaded by [York University Libraries] at 11:14 23 November 2014

2) . . . attempted to discuss sex

5) . . . unwanted sexual attention
6) . . . staring, leering at you
7) . . . attempts to establish a sexual
relationship
10) . . . repeated requests for drinks, dinner,
despite rejection
13) . . . touching in a way that made you feel
uncomfortable
14) . . . attempts to stroke or fondle
11) . . . subtly bribed you
12) . . . subtly threatened you
17) . . . made it necessary to cooperate to be
well treated
18) . . . made you afraid of poor treatment if
you didn't cooperate
19) . . . experienced consequences for refusing

ment in organizations. The results suggest that we have made considerable

progress in meeting that goal. In particular, the three-factor structural
model appears remarkably stable. With respect to the SEQ itself, it appears
to tap the construct in a reliable and efficient manner, and the correlations
of its scales with other organizationally relevant variables support its
predictive validity and practical value (Schneider & Swan, 1994; Zickar,
1994). In the following section, we present a more thorough analysis of our
structural and psychometric results, as well as their implications and
applications; briefly review the validity data on the SEQ; and conclude with
a discussion of the problems and issues that remain to be resolved when
assessing sexual harassment.

The Structure of Sexual Harassment

We begin our remarks at the point where all discussions of applied

measurement must begin -with an analysis of the construct to be assessed -
and submit that the neglect of this analysis is arguably the single greatest
 MEASURING SEXUAL HARASSMENT 437

TABLE 4
LISREL Estimates (WLS) of the Invariant Factor Loading Matrix of the SEQ-W
- -- -

Unwanted
Gender Sexual Sexual
Harassment Attention Coercion
Downloaded by [York University Libraries] at 11:14 23 November 2014

Notes. Items are arranged in the matrix in the order that they appear in the inventory. GH
= Gender harassment. USA = Unwanted sexual attention. SC = Sexual coercion.

weakness plaguing the body of sexual harassment prevalence literature as it

exists today. Viewed from a certain perspective, such neglect is understand-
able; harassment research arose not as a matter of academic or theoretical
interest, but in response to the need to solve a pressing social problem. Early
prevalence studies tended to make a brief examination of the meager legal
and regulatory guidance that existed at that time and then got on with the
more pressing business of documenting the scope of the problem, often in
the face of considerable skepticism that a problem existed at all.
This leap into data collection, however, lacking a careful specification of
exactly what was being measured, has led to a number of unfortunate
problems that, as the field matures, we can no longer afford to ignore. Of
these problems, the failure to articulate the linkage between the legal and
operational definitions of harassment and the failure to define the dimen-
sions of its domain appear to be the most significant. We submit that the
three-factor model described here represents a promising solution to these
problems. Not only is the relation between harassment as a psychological
construct and the parallel statutory dimensions clearly articulated, but those
dimensions appear to be both necessary and sufficient to capture the
variance in sexual harassment in both organizational and educational
environments. Similarly, the dimensions appear to be cross-culturally
 TABLE 5
GFI Measures of the Three-Factor Model
- - -

Memure Value
Chi-square 133.67
Degrees of freedom 116.00
Ration of chi-square/degrees of freedom 1.15
GFI index .983
RMSR .207
Downloaded by [York University Libraries] at 11:14 23 November 2014

generalizable (at least to the degree examined so far) and stable across time,
as two of the original SEQ samples we (Gelfand et al., 1995) analyzed were
collected nearly 10 years before those on which the present analysis is based.
Examining the model from a legal perspective, it is important to reiterate
that it does not speak to the conditions under which the three categories of
behavior (gender harassment, unwanted sexual attention, sexual coercion)
become harassment in the statutory meaning of that term. Such determi-
nations can only be made in light of the totality of the circumstances
involved in any particular situation (EEOC, 1990), circumstances that tend
to turn on more subjective issues of seriousness, welcomeness, and the like.
As a practical matter, a hostile environment claim (gender harassment and
unwanted sexual attention) typically requires a showing of a pattern of
offensive behavior, whereas in quid pro quo harassment (i.e., sexual
coercion) a single instance is generally sufficient to trigger Title VII and
parallel state statutes. Even the requirement that behavior be repeated and
patterned is not invariable, however, depending on other circumstances; for
example, the EEOC noted "a single, unusually severe incident of harass-
ment may be sufficient to constitute a Title VII violation . . . [in particular]
a single unwelcome physical advance can seriously poison the victim's
working environment" (pp. 16-17). Similarly, not all instances of gender
harassment are sexual in nature; although still potentially actionable under
Title VII as impermissible sex discrimination (EEOC, 1993), such instances
may not aIways qualify as sexual harassment in the legal meaning of the
phrase. Psychologically, however, such distinctions are less meaningful, as
sexual and nonsexual gender harassment tend to co-occur, as do the three
categories more generaIly, and to produce the same psychological effects
(Schneider & Swan, 1994). The advantage of our model is that it articulates
the relation between the legal and psychological constructs without in any
sense equating them.
From a behavioral perspective, the three categories appear to be both
parsimonious and comprehensive, that is, necessary and sufficient. Al-
though it is possible to make finer logical distinctions (e.g., Till, 1980), it is
unclear that there is an advantage in doing so. For example, Till (1980)
 MEASURING SEXUAL HARASSMENT 439

distinguished between bribery and threat, but it is clear that a statement

such as, "If you sleep with me, I'll promote you," logically implies and
subsumes its reverse. On the other hand, Gruber's (1992) categories conflate
the important legal distinction between quid pro quo and hostile environ-
ment (e.g., his verbal request category contains instances of both sexual
coercion and unwanted sexual attention). Thus, from both a logical and
empirical perspective, we propose that our model provides a parsimonious
yet comprehensive answer to the question, "What is sexual harassment?"
Downloaded by [York University Libraries] at 11:14 23 November 2014

Harassment as a Construct:
Locating the Psychological Process

In addition to its purely structural and operational definitions (i.e., multiple

items that correlate with one another in a particular fashion, as well as with
other variables in theoretically expected ways), the notion of a construct
presupposes that the indicators covary by virtue of their relation to some
psychological process. To take a simple example, we would expect items
and facets of an arithmetic test (i.e., addition, subtraction, multiplication,
and division) to "hang together," that is, covary, because answers to items
assessing those facets are a function of the underlying cognitive process of
arithmetic reasoning.
Of course, our analogy is not exact; the process of arithmetic reasoning
is located within the individual who produces the responses on the
correlated items of the arithmetic test, whereas sexual harassment resides in
the extra-individual experiences of the victim. Thus, what analogous
theoretical process can be proposed to account for observable interitem
correlations in this case? We suggest two complementary, contemporaneous
processes-one organizational and the other person~logical.~ First, it is
relatively clear by now that sexual harassment arises largely from organi-
zational conditions that facilitate its existence, not from individual deviance
(Culbertson, Rodgers, & Rosenfeld, 1994; Fitzgerald, Hulin, & Drasgow,
1995; Gelfand & Drasgow, 1994; Pryor, LaVite, & Stoller, 1993; Zickar,
1994); thus, it seems reasonable to argue that behaviors covary within and
between our conceptual facets because the organizational norms that
govern them are similar. For example, Hulin (1993) and Hulin, Fitzgerald,
and Drasgow (in press) identified three aspects of organizational climate
that have been shown to give rise to a greater incidence of sexual
harassment: risk to victims for complaining, the likelihood that their
complaints will not be taken seriously, and the probability that offenders
will not be sanctioned in any meaningful way. Similarly, Pryor et al. (1993)

4We are indebted to John B. Pryor, who first suggested this conceptualization to us in a
thoughtful review of an earlier version of this article.
 showed in both field and laboratory studies that lenient management norms
are related to higher levels of sexual harassment.
Thus, we argue that the organizational conditions and social norms that
permit or encourage unwelcome sexual attention apply to each of the
various aspects of such attention (e.g., leering, requests for dates, and
attempts to stroke or fondle) as well as to other aspects of harassment (i.e.,
gender harassment and sexual coercion). It follows that when norms are
weak and organizational tolerance is high, such behaviors will be not only
prevalent but will co-occur.
Downloaded by [York University Libraries] at 11:14 23 November 2014

Second, although we argue that organizational (i.e., environmental)

conditions are paramount, we also emphasize that they are not everything.
Individual differences in the propensity to harass do exist (Pryor, 1987;
Pryor & Stoller, 1994) and can be facilitated or inhibited by environmental
norms (Pryor et al., 1993). When men harass, it appears to be the case that
their behavior is multidimensional; that is, men who tell suggestive stories
are also likely to make crude sexual remarks and sexist, misogynist
putdowns, obscene nonverbal gestures, body language, and the like. Of
course, the norms and culture of the particular workplace (e.g., hospital,
university, shop floor, and coal mine) will have a great deal to do with the
form in which such harassment is expressed; this is yet another example of
the ways that organizational environments shape individual behavior. We
suggest that the combination of these organizational and individual factors
are the most promising explanations for the pattern of behaviors assessed
by the SEQ.

The Measurement of Sexual Harassment

Judged by the traditional standards of applied measurement, the SEQ

appears to be a reliable and valid measure of sexual harassment. In
addition, criterion-related validity data appear to be quite promising. For
example, higher levels of organizational tolerance (i.e., perceived climate)
are reflected in higher scores on the SEQ (Zickar, 1994); similarly, higher
SEQ scores are associated with less satisfaction with both coworkers and
supervisors (but not with work itself), as well as with organizational
withdrawal and commitment (Schneider & Swan, 1994). Such data provide
evidence of the high cost that harassment inflicts on organizations.
With respect to the price paid by individual victims, Schneider and Swan
(1994) provided one of the first empirical and objectively assessed accounts
of the relation between SEQ scores and psychological distress. Although
this relation has been discussed in the literature from its inception, virtually
all data has been provided by self-identified victims or clinical accounts (see
Gutek & Koss, 1993; Koss et al., 1994, for reviews; and Dansky &
Kilpatrick, in press, for an exception). Schneider and Swan, however,
 related scores on the SEQ to objective psychometric measures of psycho-
logical distress (e.g., anxiety, depression, and symptoms of posttraumatic
stress disorder) in an unselected sample, providing strong support for the
contention that sexual harassment is a threat to women's psychological
well-being. In addition, Gelfand and Drasgow's (1994) modeling analysis
demonstrated that SEQ scores are strongly related to physical health, both
directly and indirectly, via their impact on psychological well-being. Taken
together, the data provide convincing evidence of the validity of the SEQ.
Problems remain, however, and it is to these that we now turn.
Downloaded by [York University Libraries] at 11:14 23 November 2014

Base rates. Probably the most obvious of these problems is the

persistent issue of markedly skewed base rates on items assessing sexual
coercion and some of the more blatant forms of unwanted sexual attention.
Despite our attempts to develop more sensitive wording and scaling
procedures, base rates on these items never rose beyond 5% and, in some
cases, were considerablylower. We argue that this simply reflects reality; all
prevalence surveys to date have indicated that quid pro quo situations are
relatively rare, despite the popular stereotype that equates sexual harass-
ment with sex for promotion, retention, and the like. Although it is
probably the rare woman who has never had to "be nice" to the boss, smile,
or acquiesce gracefully to small indignities, it is clear by now that sexual
coercion in the workplace is a relatively infrequent event compared to other
forms of sexual harassment. To the degree that this is true, our base rates
simply reflect the validity of the items assessing this aspect of the construct.
Although complicating the application of various statistical techniques (and
thus the lives of researchers interested in modeling the antecedents, process,
and outcomes of this problem), this situation presents no practical problem
for conducting incidence studies, the application in which organizations are
most likely to be interested.

Scoring, The SEQ can be used to assess either individuals or organi-

zations, and the issues having to do with scoring are different for each.
With respect to organizational assessment, the process is relatively straight-
forward, involving the simple computation of frequency distributions at the
desired level of analysis (e.g., separate frequency distributions for each of
the three types, combined categories, and the like). Although the matter of
what constitutes a "case" does arise and must be addressed (e.g., Should an
individual who endorses a single gender harassment item and indicates via
the scale that this was an isolated happening be classified as sexually
harassed?), this is a practical issue that can be decided on the basis of the
organization's goals in conducting the assessment. And, the data tell us that
such isolated happenings appear to be relatively rare.
Thornier issues arise, however, when determining how to assign scores to
 442 FITZGERALD, GELFAND, DRASGOW

individuals, the situation in which researchers are more likely to be

interested. Although it is obviously possible to treat harassment as a
dichotomous variable, this seems to be a serious oversimplification. At the
very least, it would seem necessary to allow for the effects of both type and
frequency, and it is in this manner that we assigned scores for our analysis.
Each participant was assigned a score for each of the three subscales
reflecting the simple sum of within-scale responses, responses that are
weighted by the frequency with which they were experienced. Given that
harassment is a psychological stressor, however, it also seems important to
Downloaded by [York University Libraries] at 11:14 23 November 2014

capture the respondent's appraisal of the situation as benign, irrelevant, or

threatening (Lazarus & Folkman, 1984; see also Fitzgerald, Swan, & Fisher,
1995, for an in-depth discussion of harassment as an instance of psycho-
logical stress). Such appraisals invoke the issue of how to score or scale the
seriousness of a behavior or situation, a complex issue that we address in the
following section.

Seriousness. The entire issue of sexual harassment seriousness, par-

ticularly the relative importance of objective versus subjective criteria, is a
complex one that has yet to receive the attention it deserves (cf. Gruber,
1995). Although some behaviors are, in an objective sense, clearly more
serious than others (e.g., sexual assault is obviously a greater stressor than
offensive remarks), the great majority of stimulus-based severity distinc-
tions appear to be made on a more or less arbitrary basis. For example, it
is unclear on what basis the USMSPB (1981, 1987) concluded that being
deliberately touched was less serious than receiving unwanted letters and
telephone calls; similarly, although Gruber (1992) articulated a logical
rationale for his seriousness ratings, there is, at the least, room for
reasonable people to disagree. For example, it is unclear whether, as he
asserted, personal remarks made to a woman are always more offensive
than remarks made to others about her. If nothing else, it would appear to
depend on the remarks being made.
It is our belief that much of the fuzziness concerning this concept has to
do with a failure to specify exactly what is meant by seriousness, under what
circumstances, and for what purpose. From a psychological perspective, it
is clearly not possible to achieve a purely stimulus-based classification, if
only because individual differences in target (i.e., victim) history, percep-
tions, and evaluations inevitably account for so much of the variance in
reactions. However, it is also clear that certain stimulus characteristics (e.g.,
frequency) clearly contribute to the seriousness of outcomes (Schneider &
Swan, 1994). Similarly, although the law has determined that quidpro quo
harassment is generally more serious than a hostile environment, this is not
necessarily true from the perspective of victim outcomes and is not even
always true in the courtroom. Space precludes a more extensive discussion
 MEASURING SEXUAL HARASSMENT 443

of this issue here; rather, we content ourselves with noting that, for
practical purposes, the relative power and sensitivity of weighted versus
unweighted scores is an empirical question that can only be solved via
appropriate data and analysis. In this case, we included only frequency as
an empirical indicator of seriousness; however, theory and some data
suggest that other, more subjective, evaluations may, in the end, prove
superior. For example, Swan (1994) demonstrated that simple ratings of
perceived offensiveness can double the amount of variance accounted for in
outcomes. This appears to be an important direction for future research.
Downloaded by [York University Libraries] at 11:14 23 November 2014

Relationship among SEQ subscales. The final issue we raise has to

do with the substantial intercorrelation among the factors of the SEQ
(average r = .74), suggesting less differentiation than is desirable. To some
degree, this reflects a fact of life; in the real world of work, experiences of
sexual harassment are often multidimensional phenomena, with various
aspects of the problem predominating in the experience of any particular
victim. Sexual coercion, by definition, implies unwanted sexual attention
(although the reverse is not necessarily or even usually so), and gender
harassment is often expressed in sexual terms. It is also the case that the
SEQ collapses, so to speak, across different behavioral sequences and
perpetrators, thus artificially inflating the relation among the three dimen-
sions. Although these factors account to some degree for the
intercorrelation, it is also possible that the SEQ-W in its present form
provides somewhat limited coverage of the domain of gender harassment;
in particular, the blatantly hostile and often physically dangerous experi-
ences reported by women in the traditionally most male-dominated occu-
pations (e-g., coal miners; railroad workers; and some aspects of the police,
fire, and military services) are not well represented. The inclusion of such
experiences may serve to lower the correlation among the factors, although,
given that misogyny is so often expressed in sexual terms, this is an
empirical question.

CONCLUSION

This brings us full circle to the original, central argument of this article-
that sexual harassment is most appropriately conceptualized as a psycho-
logical construct. Although its dimensions are now known, its population
of observable indicators (i.e., items) is potentially infinite. As with other
constructs, some indicators may be more salient (i.e., valid) depending on
the population being assessed. This suggests the desirability of using our
theoretical framework to guide the development of a comprehensive item
bank containing large numbers of indicators from which organizationally
 444 FITZGERALD, GELFAND, DRASGOW

(and occupationally) specific versions of the SEQ can be tailored and linked
to the validity data available on the base version of the measure. This
process, common in intellective measurement but so far untried in the
present arena, promises to keep us occupied for some time to come.

ACKNOWLEDGMENTS

An earlier version of this article was presented at the Ninth Annual

Downloaded by [York University Libraries] at 11:14 23 November 2014

Conference of the Society for Industrial/Organizational Psychology, April,

1994, Nashville, TN.
We thank Chuck HuIin and the other members of the SexHarass
Research Group for their collaboration and contribution to the research
project, on which this article is based, and Larry Hubert and John B. Pryor
for their assistance in clarifying the concept of harassment as a psycholog-
ical construct.

REFERENCES

Arvey, R. D., & Cavanaugh, M. A. (1995). Using surveys to assess the prevalence of sexual
harassment: Some methodological problems. Journal of Social Issues.
Beere, C. A. (1990). Gender roles: A handbook of tests and measures (Vol. 1). Westport, CT:
Greenwood.
Culbertson, A. L., Rodgers, W., & Rosenfeld, P. (1994, August). Organizational change: The
case of sexual harassment. Paper presented at the annual meeting of the American
Psychological Association, Los Angeles.
Dansky, B., & Kilpatrick, D. (in press). Effects of sexual harassment. In W. OYDonohue(Ed.),
Sexual harassment: Theory, research, and treatment. Needham Heights, MA: Allyn &
Bacon.
Fitzgerald, L. F., & Hesson-McInnis, M. (1989). The dimensions of sexual harassment: A
structural analysis. Journal of Vocational Behavior, 35, 309-326.
Fitzgerald, L. F., Hulin, C. L., & Drasgow, F. (1994). The antecedents and consequences of
sexual harassment in organizations: An integrated model. In G. P. Keita & J. J. Hurrell, Jr.
(Eds.), Job stress in a changing workforce: Investigatinggender, diversity, andfamily issues
(pp. 55-74). W a s h i i o n , DC: American Psychological Association.
Fitzgerald, L. F., & Shullman, S. L. (1985, August). The development and validation of an
objectively scored measure of sexual harassment. Paper presented at the annual meeting of
the American Psychological Association, Los Angeles.
Fitzgerald, L. F., Shullman, S. L., Bailey, N., Richards, M., Swecker, J., Gold, Y., Ormerod,
A. J., & Weitzman, L. (1988). The dimensions and extent of sexual harassment in higher
education and the workplace. Journal of Vocutional Behavior, 32, 152-175.
Fitzgerald, L. F., Swan, S., & Fischer, K. (1995). Why didn't she just report him? The
psychological and legal implications of women's responses to sexual harassment. Journal of
Social Issues, 51, 117-138.
Gelfand, M., & Drasgow, F. (1994, April). Antecedents and consequences of sexual harass-
ment in organizations. Paper presented to the Society for Industrial-Organizational
 MEASURING SEXUAL HARASSMENT 445

Psychology, Nashville, TN.

Gelfand, M. J., Fitzgerald, L. F., & Drasgow, F. (in press). The structure of sexual
harassment: A confirmatory analysis across cultures and settings. Journal of Vocational
Behavior.
Gruber, J. E. (1992). A typology of personal and environmental sexual harassment: Research
and policy implications for the 1990's. Sex Roles, 26, 447-464.
Gutek, B., & Koss, M. P. (1993). Changed women and changed organizations: Consequences
of and coping with sexual harassment. Journal of Vocational Behavior, 42, 28-48.
Hulin, C. L. (1993, May). A framework for the study of sexual harassment in organizations:
Climate, stressors, andpatterned responses. Paper presented at the annual meetings of the
Society for Industrial/Organizational Psychology, San Francisco.
Downloaded by [York University Libraries] at 11:14 23 November 2014

Hulin, C. L., Fitzgerald, L. F., & Drasgow, F. (in press). Organizational influences on sexual
harassment. In J. Stockdale & B. A. Gutek (Eds.), Women and work. Newbury Press, CA:
Sage.
Joreskog, K. G. (1971). Simultaneous factor analysis in several populations. Psychometrika,
36, 409-426.
Joreskog, K. G., & Sorbom, D. (1988). PRELIS: A Preprocessor for LISREL (2nd ed.).
Mooresville, IN: Scientific Software.
Koss, M. P., Goodman, L., Browne, A., Fitzgerald, L. F., Keita, G. P., & Russo, N. F.
(1994). No safe haven: Male violence against women at home, at work, and in the
community. Washington, DC: American Psychological Association.
Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal, and coping. New York: Springer.
Pryor, J. B. (1987). Sexual harassment proclivities in men. Sex Roles, 17, 269-290.
Pryor, J. B., LaVite, C. M., & Stoller, L. (1993). A social-psychological analysis of sexual
harassment: The person/situation interaction. Journal of Vocational Behavior, 42, 68-83.
Pryor, J. B., & Stoller, L. M. (1994). Social cognition processes in men who are high in the
likelihood to sexually harass.
Schneider, K., & Swan, S. (1994, April). Job-related, psychological and health-related
outcomes of sexual harassment. Paper presented at the annual conference of the Society for
Industrial/Organizational Psychology, Nashville, TN.
Swan, S. (1994). [Simpleratings of perceived offensiveness can double the amount of variance
accounted for in outcomes]. Unpublished raw data.
Till, F. (1980). Sexual harassment: A report on the sexual harassment of students. Washing-
ton, DC: National Advisory Council on Women's Educational Programs.
U.S. Equal Employment Opportunity Commission. (1980). Guidelines on discrimination
because of sex. Federal Register, 45, 74676-74677.
U.S. Equal Employment Opportunity Commission. (1993). Proposed guidelines on harass-
ment based on race, color, religion, gender, national origin, age or disability. Federal
Register, 58, 5 126-51269.
U.S. Merit Systems Protection Board. (1981). Sexual harassment in thefederalgovernment: Is
it a problem? Washin@on, DC: U.S. Government Printing Office.
U.S. Merit Systems Protection Board. (1987). Sexual harassment in the federal government:
An update. Washington, DC: U.S. Government Printing Office.
Zickar, M. (1994, April). Antecedents of sexual harassment. Paper presented at the annual
conference of the Society for Industrial/Organizational Psychology, Nashville, TN.