Nursing Education: Developing Specification Equations for Selection and Retention MARY H. HUCH, RN, PHD,*
REX L. LEONARD, PHD,“~ AND
KENNETH U. GUTSCH,
Because nursing is a professional area in which costs continue to rise faster than in other academic areas, it is imperative to provide quality education that is wlthin the financial reach of those students who can benefit from it. This study sought to develop a specification equation that could be used to predict retention in the academic area and workplace. (Index words: Nursing education; Predictors; Retention; Selection) 3 Prof Nurs 8:170-775, 7992. Copyright 0 7992 by W.B. Saunders Company
B
ECAUSEof the many and varied expenses incurred
perative
in educating
nursing
students,
to look for ways to improve
it is im-
selection
and
PHD$
formance,
frequently
sets out to predict majoring
remains attrition
in a highly specialized
or if one elects to predict educated lege,
a question.
professionals
it is essential
biographical predictive
attrition
three approaches
rates among
who have graduated
data on which
validity
highly
from col-
interviews,
content,
has been to establishing
est is predictive
if one
area, such as nursing,
to use tests,
validity
Thus,
rates for college students
and/or
concurrent,
established. validity,
or
Of these the strong-
because it mandates
that the
test, interview, or biographical data gathered be collected on a sample of the students coming into the specialty area and then correlated with specific criterion-related contingencies, eg, graduation within the
retention procedures so that those most likely to succeed are selected for admission. Attrition rates for nursing students have ranged from a low of 20 per
area of the specialty selected and/or high grades in the areas that precede graduation. Of the three approaches
cent to a high of 4 1 per cent in various national
to predicting-testing,
ies (Munro,
1980; Rosenfeld,
not an easy task to predict
stud-
1988). Althoughit is
those who will be success-
ful, there are at least three approaches through which the future performance of college students can be predicted:
test data,
interview
data, and/or
biographical
data. Naturally, predictors must be both reliable, ie, they must measure with consistency that which is to be measured, they purport
and valid, ie, they must measure to measure.
what
Although test reliability usually has been well established on standardized tests, the problem of test validity,
ie, validity
as a predictor
of successful
per-
*Professor of Nursing, University of Southern Mississippi, Hattiesburg, MS. tProfessor of Educational Leadership and Research, University of Southern Mississippi, Hattiesburg, MS. $Professor of Counseling Psychology, University of Southern Mississippi, Hattiesburg, MS. Phase I of this study was funded by grants from the University of Southern Mississippi Department of Research and Sponsored Programs and the Gamma Lambda Chapter of Sigma Theta Tau. Address correspondence and reprint requests to Dr Huch: University of Souther Mississippi, Southern Station. Box 5095, Hattiesburg, MS 39406-5095. Copyright 0 1992 by W.B. Saunders Company 8755-7223/92/0803-0009$03.00/O
170
biographical
data-the
interviewing,
and gathering
best predictors
of performance
for most jobs have been schenbaum & Weisberg,
tests (Howell, 1976; Kir1990). Among the tests
most frequently used to determine potential job performance are aptitude and ability tests. These are sometimes recognized as “maximum performance” tests because they are designed to tease out the maximum performance of those who take them. Yet, in the plethora of studies sponsored by the National League for Nursing in which researchers deal with attrition (Levine, 1957; Saleh, Lee, 81 Prier-r, 1965; Schwirian,
1977,
1978a,
on the use of maximum at best, quite poor.
1978b),
predictions
performance
based
tests have been,
Purpose This study sought to develop a specification equation that could be used to predict retention of nursing students in the selected program of study. Selected
Literature
The prediction studies found in the nursing literature focus on a variety of variables ranging from stan-
Journal of Professional Nursing,
Vol 8, No 3 (May-June),
1992: pp 170-175
DEVELOPING
dardized
SPECIFICATION
tests to grade point
that can be used for prediction points
in the student’s
McKinney, Sharp, 1986;
(Glick, Small,
1984).
The variables
are available
Data available
was used to predict McClelland,
O’Dell
at varying success in
& Young,
& Coonrod,
1986;
1988;
Other
researchers & Whitley,
1985) used data
before entry
into nursing
courses and there-
after. Yet another end of a program
(Payne
and
Krupa,
Quick,
available
averages.
matriculation.
at the time of admission some studies
171
EQUATIONS
time of data availability of study (Breyer,
& Duffy,
capacity”
programs,
provide quality all students Cattell,
will eventually
education
within
Eber, and Tatsouka
jective formulas for prediction
(1970)
linked to personality
posit that ob-
tests can be used
rather than merely relying
Furthermore,
they contend
be developed
for a given occupational
is at the
on intuition.
that a typical
profile
can
group.
Problem
1984). In seeking
Nursing is a professional area in which educational costs continue to rise faster than they do in other academic areas.
evidence
bearing
among
directly
nursing
used are per-
sonality tests. They are sometimes referred to as “typmeasures because they do not necical performance” essarily challenge the person to perform at a maxi-
this study
into two phases, ie, phase I and phase II,
and sought
evidence
1.
regarding
Phase I: Do personality who drop
three questions. profiles
out of a nursing
of students
program
differ
from the program?
2. Phase II*: Once graduated less frequently
on the high
students,
was divided
from those who graduate the measures
to
reach of
who desire it.
rate of attrition
Among
find it difficult
the financial
from the school of
nursing, do personality profiles of those students who become long-work tenure nurses, ie, nurses who stay in the profession after
mum level. Unless they are used in creative ways, the
receiving an undergraduate degree, from those who become short-work
results of these tests have no specific outcome contingency against which measurement means anything.
nurses, ie, those who drop out of the nursing profession after they graduate from college as
However, when used for predicting they seem to have practical value and now seem to have potential for the
nurses?
study of attrition among ple, in the Sloat, Gutsch, which
Sixteen
nursing students. For examand Leonard study (1983) in
Personality
Factor
Questionnaire
(16PF) profiles of drug users were compared with those of nonusers, they used Y as the predicted discriminate score and X as the 16PF sten score for each personality factor. Ultimately, they found nine factors that discriminated between the two groups. They then constructed an extrapolation table based on discriminate weights to show movement toward or away from drug use. It was their contention that the statistical procedures incorporated in studies they did with drug users versus nonusers and alcoholics versus nondrinkers indicated substantial support for the idea that when clear-cut behavior was defined as a criterion against which personality measurements could be made, the specification equation they designed could be used for prediction. Furthermore, it was their contention that similar equations could be developed to predict potential dropouts in nursing. Nursing is a professional area in which educational costs continue to rise faster than they do in other academic areas. Thus, nursing programs, as “limited
3. Phases I and II: When applying
differ tenure
the statistical
techniques developed by Sloat, Gutsch, and Leonard (1983): (a) Can the retention of nursing students be predicted more accurately than it has been in the past? (b) Can longtenure nurses be identified through cessing of personality test results?
the pro-
Procedure SAMPLE
Subjects for this longitudinal study consisted of 15 1 nursing majors at a southern public university who agreed to participate. All entering students during 1983 to 1984 were approached to participate in the study. Those who agreed to participate signed a consent form after the purpose of the study was fully *Long-work tenure versus short-work tenure comparisons will be made at a later date when the emphasis for the study will be on phase II of the research, and will include the characteristics that differentiate between nurses who stayed in the profession of nursing once they received their undergraduate degree and those who dropped out after entering the field.
HUCH, LEONARD, AND GUTSCH
172
explained.
No special inducements
to participate
were
coursework.
Stepwise
was then
provided. Subjects
were divided
into
groups
phase I based on successful completion of study (graduates, nursing
program
the university
N =
(transfers,
tinuation
N =
completely
of studies
of the program
121), subjects
and transferred
left the university
at the end of who left the
to another
major in
16), and subjects and reported
(inactives,
N =
who
noncon-
14).
differences
that
could
nursing
differentiate
students
graduation
of specification between
who remained
(1) undergraduate until
out before gradu-
of the program
of nurs-
been used in over 2,000 reported research studies. At least 15,000 adults were tested in the latest standardization of the instrument. Factor analysis was used to establish construct validity of the instrument, which yielded 16 personality factors. Scale reliabilities (testretest) for the 16PF range from .6 1 to .9 1 across all Equivalence
coefficients
be-
tween the various forms taken pairwise range from .15 to .82 across each scale trait (Cattell, Eber, & Tatsouka, 1970). DESIGN
Once data were collected,
estab-
lishes the best possible discrimination
using a forward variables
procedure.
At each step,
were
added only when they made the best contribution discrimination
to
in the presence of those already in the
model.
Then all of those variables
tested,
and the ones that were deleted
a data
ing who rendered short-tenure service (s2 years) to their profession and those who rendered long-tenure service (>2 years) to their profession after receiving their undergraduate degree in nursing. The 16PF has
forms of the instrument.
This technique
selection
in the model were
did not make
significant
at the conclusion
of each
Step 2
equations
in the program
and those who dropped
ation and (2) undergraduates
analysis
step.
Form A of the 16PF was used to establish base for the development
discriminate
a data base from which
could be measured.
contributions INSTRUMENT
multiple
used to establish
raw scores for all groups
The structure tions
between
function,
coefficients,
which
each variable
and
are the correlathe discriminant
were used to help select the principal
factors
of the 16PF. Only those personality factors that remained in the final discriminant model and had the largest structure coefficients were used for study. These factors, which discriminated between groups, became the data base from which predictions were made.
Using
Y as the predicted
discriminate
score
and X as the sten score for each personality factor, the equation for prediction was constructed as follows: Y = WiXi where Wi is the ich discriminant
function
weight.
Group membership, ie, those who stayed versus those who dropped out, was determined by the proximity of a predicted score to the centroid of that group. In other words, if the two groups were of equal size, half the linear distance between the two centroids would become a cutting score for prediction preference
of a
were converted to stens, and comparisons were made for phase I of the study. This included a 5-year study
student to initially undertake a study of nursing dicates some predisposition to success, a weighted
inav-
to determine the differences between those subjects who graduated with an undergraduate major in nursing and those who dropped out of nursing before re-
erage of the group centroids discriminant cutting score.
ceiving their undergraduate degree. Comparisons for phase I of the study, ie, looking at students who graduated from nursing school versus those who dropped out, was done through the use of a two-step design. Step 1
Univariate F tests on each of the 16PF factors were used to test for significance between the students who graduated and those who became inactive, ie, those who dropped out, or transferred, ie, those who transferred out of nursing and into another area of academic
purposes.
However,
since the personal
provided
a more optimal
Findings Discriminant analyses results, which are shown in Tables 1, 2 and 3, showed factors with significant discriminations for the comparisons mentioned. These factors were the ones chosen for the discriminant function through stepwise discriminant analysis. Discussions about each table or comparison focus on those factors having both significance as well as higher structure coefficients, ie, correlations between the factor and the discriminant function (Hair, Anderson, & Tatham, 1987).
DEVELOPING SPECIFICATION EQUATIONS
1.
TABLE
173
Average Sten Scores of Significant Factors and Structure Coefficients of G Versus T Students
3.
TABLE
Average Sten Scores of Significant Factors and Structure Coefficients of G Versus T and I Students
Sten Score
Sten Score Structure Coefficient
G
Factor
(N = 21)
(N =’ 16)
I3 C E
1.8 5.4 5.7 5.5 1.9
2.0 4.4 4.3 6.6 2.0
I Q4
A comparison
of personality
of the school of nursing those students
factors of graduates
with inactive
who stopped
school of nursing,
,336 ,433 ,610 - ,482 174
attending
and transfer
(G)
(I) subjects,
ie, those
students who transferred out of the school of nursing to other academic areas of the university, showed the following results. The profiles of G subjects subjects
differed
from
I or T
scored significantly
tors E and C and significantly
cients of the greatest
did I subjects
I L M 0
Q3 Q,
terested
(Table
-
career and wanted
to know who
the standardized sten score for each personality factor in Table 3, computer analysis yields the following classification
student file--F
equation:
of the equation
proceeds
advisor intended
Suppose
to work with entering
fication
in-
Average Sten Scores of Significant Factors and Structure Coefficients of G Versus I Students Sten Score G (N = 121)
5.6 70 6.0 5.5 4.4 5.7 5.9 61
(N :
14)
4.0 6.4 6.1 6.6 3.6 6.2 5.2 5.3
best done by first standard-
Structure Coefficient ,459 192 - ,025 - ,369 ,256 -.I43 -.014 ,287
by constructing
the profile are into the classi-
a table
(Table
4).
is used because results from discrimie, discriminant scores and coefficients,
are more convenient to plot when standardized. On Table 4 the adviser first locates factor F on the left side and follows the factor F row across the table until value
2.
equation
Standardization inant analysis,
that an
freshmen
= 4, Q4 = 3-is
izing scores. Once the scores forming standardized, they can be substituted
the development
as follows.
these factors to the profile of an entering
with the following 16PF personality pro= 2, G = 3, I = 7, L = 6, M = 7,0 =
2, Qj
2).
Once the results were determined,
Q3 Q,
in a nursing
Applying
magnitude.
Qi (Table 3) than did the combined group of dropouts, ie, T plus inactive subjects.
I
-
,314 ,176 ,481 042 ,331 ,148 ,282 ,269
had potential to complete the program of study. Denoting Y as the predicted discriminant score and X as
lower on Factor
3. G subjects scored significantly higher on factors F, M, and Q4 and lower on factors I and
M 0
4.8 6.6 66 5.4 3.6 6.1 7.6 5.5
Structure Coefficient
Y = (.398)X, + (.529)X, - (.560)X, - (.251)X, + (.528)X, - (.378)X, - (.551)X,, + (.432)X,,.
2. G subjects scored significantly higher on factor F and significantly lower on factor I than
F G H
56 70 5.5 5.3 4.4 5.7 6.9 6.1
F G
higher on fac-
I of the 16PF than did T subjects (Table 1). These factors also showed structure coeffi-
Factor
T and I (N = 30)
in three areas:
1. G subjects
TABLE
G (N = 121)
ie,
classes in the
(T) subjects,
Factor
the column
under
of - .608. This
discriminant
weight
sten 2 is reached value
to find the
is the product
.398 times the standardized
of the sten
of - 1.528 and thus becomes the contribution of factor F in the calculation of Y. Similarly, the adviser continues this process of gathering the products for each of the predictors. Once assembled, products are placed into the equation and substituted for the student’s original sten profile: Y = -.608 - .896 - .372 - .085 + .733 + .654 + .875 - .683. Adding the positive numbers yields a sum of 2.262, and adding the negative numbers yields a result of - 1.961. Combining these two numbers yields a score of .301, ie, the predicted discriminant score.
174
TABLE
HUCH, LEONARD, AND GUTSCH
Example of Standardization Respective Weights
4.
Table With Products
of Factors,
Stens, and
Sten 1
2
3
4
5
6
7
- ,785 i .a30 1.347 ,578 - .a49 ,826 1.733 -1.136
- ,608 -1.519 1.061 ,445 - ,937 ,654 1.447 -.910
- ,430 - ,896 ,774 ,313 - ,321 ,486 1.161 - ,683
- ,253 - ,896 ,487 ,180 - ,058 309 .a75 - ,456
- ,075 - ,585 201 ,047 ,206 ,137 ,589 - ,230
,102 - ,274 - ,085 - 085 ,469 - 036 ,303 - ,003
,280 ,037 - .372 -.218 733 - ,208 ,017 ,224
Factor
F G
I L M 0
Q, Q,
-
8
-
,457 ,344 ,658 ,350 997 380 269 450
9
10
,635 ,659 - 945 - 483 1.260 - ,552 - ,555 ,667
al2 970 - 1.232 -.616 1.524 - ,725 - .a41 ,904
NOTE: Factor F-low score (L): sober, taciturn, serious; high score (H): happy-go-lucky, enthusiastic, impulsive, lively. Factor G-L: expedient, disregards rules; H: conscientious, perservering, moralistic. Factor I-L: tough-minded, self-reliant, realistic H: tender-minded, intuitive, unrealistic. Factor L-L: trusting, adaptable, free of jealousy; H: suspicious, self-opinionated, hard to fool. Factor M-L: practical, careful, conventional; H: imaginative, careless of practical matters. Factor O-L: unperturbed, self-assured, confident. H: apprehensive, self-reproaching, worrying. Factor Q,-L: undisciplined self-conflict, careless of protocol, H: controlled, socially precise, compulsive. Factor Q,-L: relaxed, tranquil, torpid; H: tense, frustrated, driven. From Cattell, R. B., Eber, H. W., & Tatsouka, M. M., Handbook for the Sixteen Personality Factor Quesfionnaire (l.PF), copyright 1986 by the Institute for Personality and Ability Testing, Inc Reproduced by permission.
To develop the meaning now needs to establish two centroids
of such a score, the adviser
a cutting
within
score between
the two groups.
for this data were .233 and - .940. represent the averages of discriminant
the
The centroids These points scores for the
two groups. The cutting score is the criterion score against which each individual’s discriminant score is judged
to determine
into which group the individual
ample,
both groups,
ie, G versus T and I, were com-
posed of people (primarily
and I subjects situations rewarded.
were probably
in which
In reviewing
significant
as more “down-to-earth,”
=
+
realistic,
N42
3OC.233) + (121) (-
.940)
121 + 30 =
6.99
-
113.74
151
=
- .707.
Note that Z, is the centroid for group 1, and N, is the number of students in group 1. Computation of this equation shows that the sample profile score is greater than - .707 and thus falls into the graduation group. Knowing this, the adviser can discuss entrance into the nursing program from a more specific perspective.
Discussion In synthesizing this material, it appears that when comparing G with T and I subjects some common characteristics reflected well on both groups. For ex-
differences
between
were G and
patient, general,
lower on factor I could best be described
tough
jects tended
NI + Nz
to seek
behaviors
I subjects, (Table 2), one might conclude that the G subjects who were significantly higher on factor F and
score: N2Z1
who might
more inclined
these particular
significantly
=
young women)
well be described as enterprising, incisive, and resilient in their relationship with others. However, the T
should be classified. Because the groups were not of equal size, the weighted average of the group centroids is then calculated to provide an optimal cutting
z,
1970,
minded,
independent, and responsible,
to be more demanding,
self-reliant, and I sub-
dependent,
im-
temperamental, unrealistic, and fussy. In I subjects seemed less capable of becoming
team players. When reviewing differences between G & T subjects, (Table 1) it becomes obvious that the G subjects scored significantly higher on factors E and C and significantly lower on Factor I of the 16PF. From these differences one might conclude that G subjects were probably more mature, stable, calm, assertive, self-assured, self-reliant, and realistic; that they were no-nonsense people with a tough minded, competitive, but down-to-earth approach to life. On the other hand, T subjects were more dependent, docile, submissive, and easily led. They seemed to be more impatient and temperamental, more easily annoyed, and more easily frustrated than were those who stayed in the program.
DEVELOPING
SPECIFICATION
Interestingly
enough,
low on factor students
EQUATIONS
both G and I subjects
M, which
could
who go into nursing
over details,
conventional
tical matters,
175
indicate
are probably
thinkers,
many
concerned
attentive
and show good stability
scored
that
to prac-
during
emer-
gencies. Perhaps
the
greatest
was reflected to be lively,
and cheerful, mistic,
the personality
students
to understand
Baccalaureate centrated
divergence
in personality
on factor F where G subjects enthusiastic,
and T and I subjects
sober, restrained,
reticent,
talkative,
carefree,
were more pessiintrospective,
and
students lem.
to begin
families,
money
data such as has been discussed
can be
used by an adviser to guide students using something more than intuition as the student makes career choices. dent
Personality
and thus
traits
relate to needs of the stu-
advisement
can occur based on hard
is spent
Attrition
of their
and other sources in By identi-
who are more likely to transfer to the resources of the
can be saved and directed
toward
may have a greater
success. Additionally,
the
is a costly prob-
education.
major or become inactive,
an area
likelihood
for
the resources of the school can
be put to more effective use in concentrating most likely to persist. Although using equation will not remove all uncertainty, vide an objective
are con-
by the students,
to secure a nursing
where the student Gathering
courses.
the school of nursing,
attempting
student
nursing
from the school of nursing
Much
another
Implications
of study
which means the stu-
2 years of study before having
fying those students
deliberate.
better.
programs
in the upper division,
opportunity
test data can assist
more clearly as well as the
themselves
nursing
dent has completed
makeup seemed
data. Furthermore,
advisers to know students
measure
for student
on those
a prediction it does pro-
selection.
References Breyer, F. J. (1984). The comprehensive nursing achievement test as a predictor of performance on the NCLEX-RN. NzlrJing and Health Care, 5, 193- 195. Catell, R. B., Eber, H. W., & Tatsouka, M. M. ( 1970). Handbook for the sixteenpersonalityfactor questionnaire (16PF). Champaign, IL: Institute for Personality and Ability Testing. Glick, 0. J., McClelland, E., & Young, J. C. (1986). NCLEX-RN: Predicting the performance of graduates of an integrated baccalaureate nursing program. Journal of Professional Nursing. 2, 98- 103. Hair, J. F., Anderson, R. E., & Tatham, R. L. (1987). Mu&variate duta analysis with readings (2nd ed.). New York: Macmillan. Howell, W. C. (1976). Essentials of industria/ and organization&psychology. Homewood, IL: Dotsey. Kirschenbaum, A., & Weisberg, J. (1990). Predicting worker turnover: An assessment of intent on actual separations, Human Relations, 43(9), 829-847.
Levine, E. (1957). Turnover among nursing personnel in general hospitals. Hospitals, 3 1, 5 1. McKinney, J., Small, S., O’Dell, N., & Coonrod, B. A. (1988). Identification of success for the NCLEX and students at risk for NCLEX failure in a baccalaureate nursing program. Journal of ProfarsionalNursing, 4, 55-59. Munro, B. (1980). Drop-outs from nursing education: A path analysis. Nursing Research, 29, 37 l-377. Payne, M. A., & Duffy, M. A. (1986). An investigation of the predictability of NCLEX scores of BSN graduates using academic predictors. Journal ofProfessionalNursing. 2 I 326-332.
Quick, N. M., Krupa, K. C., & Whitley, T. W. (1985). Using admission data to predict success on the NCLEX-RN in a baccalaureate program. Journal of Professional Nursing, 1, 364-367. Rosenfeld, P. (1988). Measuring student retention: national analysis, Nursing and Health Care, 9, 199-202.
A
Saleh, S. D., Lee, R. J., & Prien, E. P. (1965). Why nurses leave their jobs: An analysis of female turnover. Personnel Administration. 28, 25-28. Schwirian, P. M. (1977). Prediction of successful nursing performance. Parts I & II-U.S. Health Resources Administration, Nursing Division. (DHEW Publication No. HRA 77-27). Washington, DC: U.S. Government Printing Office. Schwirian, P. M. (1978a). of nurses: A multidimensional 27, 347-35 1.
Evaluating approach.
the performance Nursing Research,
Schwirian, P. M. (1978b). Prediction of successfulnursing pwformunce. Part III-Evaluation and prediction of performance of recent graduates. (Report #783970). Columbus, OH: The Ohio State University Research Foundations. Sharp, T. G. (1984). An analysis of the relationship of seven selected variables to state board test pool examinations performance of the University of Tennessee, Knoxville College of Nursing. Journal of Nursing Education, 23, 57-64. Sloat, D., Leonard, R., & Gutsch, K. (1983). Discriminant analysis for measuring psychotherapeutic change. Measurementand Evaluation in Guidance, I6( l), 36-42.