ERIC EJ750816: Some Determinants of Changes in Preference over Time

2006·0.17 MB·English

by  ERIC

JOURNALOFAPPLIEDBEHAVIORANALYSIS 2006, 39, 189–202 NUMBER2 (SUMMER2006)
SOME DETERMINANTS OF CHANGES IN
PREFERENCE OVER TIME
GREGORY P. HANLEY
UNIVERSITYOFKANSAS
BRIAN A. IWATA
UNIVERSITYOFFLORIDA
AND
EILEEN M. ROSCOE
NEWENGLANDCENTERFORCHILDREN
Resultsoflongitudinalstudiessuggestthatthestabilityofpreferences variesacrossindividuals,
although it is unclear what variables account for these differences. We extended this work by
conductingperiodic assessments ofpreference forleisure activitiesover 3to6months with10
adults with developmental disabilities. Although previous research has collectively shown that
preferences identified via repeated assessment are highly variable, our results showed that
preferenceswererelativelystableforthemajority(80%)ofparticipants.Inanattempttoidentify
some environmental determinants of shifts in preference, weprovided extended daily access to
high-preferenceitems(preference-weakeningmanipulation)andpairedaccesstolow-preference
itemswithsocialandedibleputativereinforcers duringbriefsessions(preference-strengthening
manipulation).Preferenceassessmentscontinuedoverthecourseofthesemanipulationswith2
participants. Results showed that changes in preference across time could be produced
systematically and suggest that naturally occurring changes in establishing operations or
conditioning histories contribute to temporal shifts in preference. Implications for preference
assessments, reinforcer usage,andplanned attempts tochange preferences are discussed.
DESCRIPTORS: conditioning, developmental disabilities, imposed variability, longitudi-
nalassessment, preference assessment, preference stability, satiation
_______________________________________________________________________________
The selection of reinforcers is an important et al., 1992; Roane, Vollmer, Ringdahl, &
process in the development of both skill- Marcus, 1998); as a result, various indexes of
acquisition and behavior-management pro- preference(e.g.,approach,selection,durationof
grams for individuals with severe disabilities. engagement) are now used as the basis for
Pace, Ivancic, Edwards, Iwata, and Page (1985) selecting reinforcers during treatment as well as
illustrated a two-step model in which the for arranging leisure-activity schedules.
outcomes of empirical preference assessments Assuming that initial success is observed
were predictive of subsequent reinforcement during intervention, subsequent fluctuations or
effects. These results have been replicated many decrements in performance may occur due to
timesacrossseveraldifferenttypesofassessment changes in preference over time as well as other
techniques(e.g.,DeLeon&Iwata,1996;Fisher factors. Mason, McGee, Farmer-Dougan, and
Risley (1989) conducted two preference assess-
ments approximately 1 month apart with 3
Thisresearchwassupportedinpartbyagrantfromthe
young boys with autism; Carr, Nicholson, and
FloridaDepartment ofChildren andFamilies.
Reprints may be obtained from Gregory P. Hanley, Higbee (2000) conducted eight assessments
Applied Behavioral Science Department, 1000 Sunnyside over a 1-month period with 3 young children
Ave., University of Kansas, Lawrence, Kansas 66045
with autism; and Zhou, Iwata, Goff, and Shore
(e-mail: [email protected]).
doi:10.1901/jaba.2006.163-04 (2001) examined the stability of preference in
189
190 GREGORY P. HANLEY et al.
22 adults with severe developmental disa- although positive correlations were obtained
bilities, with assessments spaced approximately for 16 of the 22 participants in the Zhou et al.
16 months apart. The results were difficult to study,onlysixofthesecorrelationsexceededthe
compare because assessments were conducted critical r value. Taken together, the results of
using different procedures, and results were these studies indicate that preferences were
reported in somewhat different ways. Mason et stable (i.e., measures on two or more assess-
al.listedapproachpercentagesinatable,Carret ments yielded highly similar results) for only 8
al. presented bar charts showing item rankings of a total of 28 participants (29%).
across all eight assessments, and Zhou et al. Thus, the literature suggests that preference,
presented bar charts of item rankings for 2 in addition to being idiosyncratic initially, may
participantswhosedatashowedahighdegreeof be relatively unstable across time. Although
stability and instability. Zhou et al. also only a small amount of data is available, it
calculated Spearman rank order correlations appears that the lack of stability does not seem
for each of the 22 pairs of assessments. to be related to the age or disability of
To provide a comparable basis for sum- participants, the type of items included in the
marizing results from the three studies, we assessments, or the length of time between
(a) assigned ranks to the items in the Mason assessments. Interestingly, the largest propor-
et al. (1989) study based on the percentage tion of participants whose preference patterns
of trials on which the items were selected, appeared to be stable was in the Carr et al.
(b) calculated rank order correlations between (2000) study, in which the paired-stimulus
rankings in the initial and subsequent assess- assessment (Fisher et al., 1992) was used.
ments (Mason et al.) or among rankings across Because only 3 individuals participated in the
assessments (Carr et al.), and (c) compared the Carr et al. study, the initial part of the current
obtained correlation coefficients in all three investigationinvolvedanextensionofCarretal.
studies to a critical r value (Salkind, 2001) of through repeated assessment of preference for
.58. We selected .58 as a guideline because it leisure items using the paired-stimulus format
was derived from a statistical textbook (Sal- across varying lengths of time.
kind), it produced reasonable agreement when Although results from repeated assessments
comparedtovisualinspectionofthedata,andit provide a descriptive analysis of the stability of
was similar to Cicchetti and Sparrow’s (1981) preference, a determination of why preferences
cutoff of r 5 .6 for a good test–retest reliability may or may not remain stable over time
coefficient. The resulting coefficients for par- requires analysis of a different sort. Time per
ticipantsintheMasonetal.studywere.26,.49, se does not produce a change in preference but
and .48, showing small or moderate positive is only correlated with other events that an
correlations between the first and second individual experiences. Moreover, we assume
assessments (none of the correlations exceeded thatvariabilityinpreferenceismorelikelytobe
the critical value). Correlation coefficients imposed by these environmental events rather
between each assessment and every other thanbeinganinherentcharacteristicofbehavior
assessment were averaged to yield correlations (Sidman, 1960). Thus, changes in preference
of .14,.64, and .75 forthe3 participants in the among activities could result from the acquisi-
Carr et al. study. Two of the three correlations tionofnewskillswithrespecttothoseactivities,
exceeded the critical value, suggesting that their intervening conditioning histories in which
preferences were relatively stable. This outcome particular items have been paired with other
is consistent with the authors’ conclusion based reinforcing events, or temporary changes in the
on visual inspection of the data. Finally, reinforcing characteristics of particular items
PREFERENCE STABILITY 191
due to their repeated presentation (i.e., satia- Amari, 1996) were conducted with participants
tion) or continued absence (i.e., deprivation) or staff members who worked with participants
(Hanley, Iwata, & Lindberg, 1999; Hanley, toidentifyitemslikelytobepreferred.Thenine
Iwata, Roscoe, Thompson, & Lindberg, 2003; items ranked highest by a respondent were
Klatt, Sherman, & Sheldon, 2000; Vollmer & includedinrepeatedpreferenceassessmentsthat
Iwata, 1991). were conducted in a paired-stimulus format
Vollmer and Iwata (1991) provided an early (Fisheretal.,1992).Eachofthenineitemswas
demonstration that the efficacy of reinforcers paired once with every other item during each
could be influenced by deprivation and satia- assessment (resulting in 36 trials per assess-
tion conditions immediately prior to sessions. ment). An approach response, defined as
More recently, Hanley et al. (2003) showed reaching towards one of the items, was scored
that pairing one item with other reinforcing during each trial (approaching both items
events could elevate its preference relative to simultaneously was blocked). The participant
another item that was not paired with other was given 10- to 15-s access to each item
events. The second part of the current study following an approach response. If no item was
extendedthefindingsofVollmerandIwataand approached on a given trial, both items were
Hanley et al. by combining elements of these describedtotheparticipant,andanewtrialwas
prior studies and examining the effects of initiated. Proportions were calculated for each
satiation and conditioning procedures in item by dividing the number of trials on which
a broader context involving multiple stimuli, an item was approached by the total number
longer delays between satiation and condition- of trials on which that item was presented.
ing operations and the assessment of their The items then were ranked inversely to their
effects, and repeated preference assessments. proportions (e.g., the item associated with the
highest proportion received a rank of 1; the
item associated with the lowest proportion
METHOD
received a rank of 9). Items that were selected
Setting and Participants on an equal number of trials were assigned an
Ten adults ranging in age from 26 to average rank (e.g., if two items were both
62 years participated. All had been diagnosed selected on eight of nine trials, both would be
with a developmental delay (mild to severe), assigned a rank of 1.5).
were ambulatory, and could follow single-step Subsequent assessments with the same items
instructions. The participants attended a work- were conducted with each participant at
shop program and were enrolled in this project approximately the same time of day across a 2-
to identify preferred activities that could be to 6-month period. The items used during the
scheduled at times when work was unavailable. preference assessments were not explicitly made
Sessions were conducted in the workshop available outside the assessment sessions, but
cafeteria or in conference rooms that contained similar items (magazines, radios) were available
tables, chairs, and, at times, other workshop in the participants’ homes throughout the
employees. All participants were given the evaluation (as is typically the case when items
opportunity to refuse to attend any session from a preference assessment are used in
(this never occurred). behavioral programs). The median number of
preference assessments conducted with each
Descriptive Assessment of Preference Stability participantwas11(range,6to16;seeTable 1).
Reinforcer assessment interviews for the The minimum number of days between assess-
severely disabled (Fisher, Piazza, Bowman, & ments was 7 for all participants, and the mean
192 GREGORY P. HANLEY et al.
Table 1
SummaryData for theDescriptive Assessment ofPreferences
Participant
Abe Bo Cathy Dana Ed Fay Gina Han Inga Jan
Totalnumberof
PAs 11 11 11 11 9 10 8 6 11 16
Numberofbaseline
PAs 11 11 11 11 9 10 8 6 4 6
Meannumberof
daysbetween
PAs 9 9.6 8.9 9.2 12.6 9.2 8.7 25 9.7 11.4
Rangeofdays
betweenPAs 7–13 7–13 7–11 7–15 7–26 7–13 7–19 8–106 7–13 7–22
Totalnumberof
daysin
assessment 90 96 89 92 101 83 70 151 97 181
PercentageofPAs
withIOA 63.6 81.8 63.6 63.6 66.7 70 50 50 63.6 68.8
MeanIOA 100 99.4 99.2 99.2 100 97.2 100 100 98.4 99.5
LowestIOAscore 94.4 94.4 94.4 94.4 88.9 94.4
Meancorrelation
betweeneachPA
andeveryother
PA(stability
measure) .28 .18 .68 .57 .65 .60 .71 .74 .70 .76
Meancorrelation
betweenfirstand
subsequent
baselinePAs
(predictive
validitymeasure) 2.11 .15 .32 .44 .49 .64 .67 .85 .77 .71
Note.Thenumbersinboldfacerepresentmeancorrelationcoefficientsthatexceededacriticalrvalue.PA
5preference assessment; IOA5 interobserver agreement.
number of days between assessments ranged assessments. These assessments served as the
from8.7to25acrossparticipants(seeTable 1). baselinesfromwhichtheeffectsofsatiationand
conditioning procedures were evaluated.
Experimental Analysis of Preference Stability The satiation procedure was applied to items
Results from several studies have indicated thatwereconsistentlyrankedthehighestduring
that variability is often evident when repeated the baselineassessments. This item wasa Walk-
preference assessments are conducted (Carr et manH radio, which Inga selected most often in
al., 2000; Mason et al., 1989; Zhou et al., all baseline assessments and Jan selected most
2001).Weexaminedwhethersimilarvariability often in six of the seven baseline assessments.
couldbeimposedonotherwisestablepreference When thesatiation procedure wasin effect, free
patterns. Toward that end, two procedures, one access totheradiowasavailabletoIngaandJan
for decreasing and one for enhancing the foraminimumof2 hrandamaximumof3 hr
reinforcing value of an item, were implemented each day, except on days in which a preference
with Inga and Jan, who were selected because assessment was conducted, when it was unavail-
stability was most evident in their repeated able. This was accomplished by providing Jan
preference assessments. Correlations between or Inga with the radio during the morning shift
preference assessments (measures of stability) and periodically checking on her to be sure it
were well above our critical r value for four of wasaccessibleandoperating(checkingbatteries,
Inga’s five assessments and for six of Jan’s seven etc.). It is important to note that there was
PREFERENCE STABILITY 193
always at least a 24-hr period between free of items, the satiation procedures were reas-
access to the radio and a preference assessment. signedtotheofficetask(Mrank53.4)andthe
During the same period in which the magazines (M rank 5 2.9), and the condition-
satiation procedure was implemented, condi- ing procedures were reassigned to the writing
tioning trials were conducted with items that pad (M rank 5 7.3) and sewing activity
wererarelyselectedduringbaselineassessments. (M rank 5 6.8) for Inga and Jan, respectively.
Bubbles (M rank 5 8.3) and a drawing book The effects of the satiation and conditioning
(M rank 5 8.3) were included in the initial procedures were determined by discontinuing
conditioning trials for Inga and Jan, respective- the procedures with the initial set of items
ly.Theconditioningprocedureinvolvedpairing (radio and bubbles for Inga; radio and drawing
these less preferred items with social reinforce- for Jan) and reassigning the procedures to new
ment (continuous attention) and consumable sets of items. Thus, an ABA reversal design was
items (diet soda and salty snack foods). The used to assess the effects of the satiation and
reinforcing efficacy of these items had been conditioning procedures with the initial set of
demonstrated in previous unrelated research items, whereas a multiple baseline design was
projects (Thompson, Iwata, Hanley, Dozier, & used to determine the effects of the same
Samaha, 2003) or was presumed based on the independent variables across sets of items.
number of requests made for these items and
RESULTS
interactions in the past. Because access to the
item to be conditioned was a requirement for Preference patternsrangedfromhighlystable
conditioning to occur and this access may to quite variable across the 10 participants.
have mitigated the effects of conditioning, an Figure 1 shows two extreme patterns. Han’s six
attemptwasmadetoequateexposuretime with assessmentsarecharacterizedbyahighdegreeof
other less preferred items. Therefore, condi- stability. For instance, the radio always retained
tioning sessions involved 15 consecutive 1-min a rank of 1 or 2, and the beading materials
trials in which one of three items was available always retained a rank of 8 or 9. Some
singly (i.e., each item was available for five variability may be noted for other items such
trials). During the bubbles (Inga) and drawing as cologne or bubbles, but the initial hierarchy
book (Jan) trials, diet soda and low-fat snack among the nine items usually held across
foods were accessible, and the therapist inter- repeated assessments. By contrast, Abe’s initial
acted and conversed with the participant six preference assessments (Figure 1, bottom)
continuously and praised all forms of engage- show a high degree of variability (only six are
ment. By contrast, when the other two low- displayedtoprovideaneasiervisualcomparison
preference items were available (drawing and with Han’s data). Six of the nine items shifted
puzzles for Inga; beads and office task for Jan), four or more ranks across six assessments. The
soda or snacks were not present, and the items ranked highest and lowest initially
therapist directed his or her attention to a task (massager and jigsaw puzzles, respectively) were
(i.e., no attention was delivered to the partic- ranked seventh and first, respectively, by the
ipant).Oneconditioningsessionwasconducted sixth assessment.
each day except on days when preference Less extreme patterns of responding were
assessments were conducted (i.e., there was more difficult to identify through visual in-
always at least a 24-hr period between a condi- spection of bar graphs, especially as the number
tioning session and a preference assessment). of assessments increased beyond six. Therefore,
Aftereffects of the satiation and conditioning rank-order correlation coefficients were calcu-
procedures had been observed with the first set lated to determine the correspondence between
194 GREGORY P. HANLEY et al.
Figure 1. The results of six consecutive preference assessments are shown for Han and Abe. The items with the
highestselectionpercentageswereassignedarankof1,andtheitemswiththelowestselectionpercentageswereassigned
arankof 9.
the each assessment and every other assessment calculated Pearson correlation coefficients for
by assigning ranks based on selection percent- each assessment comparison. Each mean corre-
ages (1 for the item with the highest selection lationcoefficientisdisplayedforallparticipants
percentage; 9 for the item with the lowest) and (including Abe and Han) in Figure 2. The
comparing the ranks from each assessment with results of each preference assessment were
every other assessment. Because Spearman rank considered to be consistent with the results of
order correlation cannot be calculated with other preference assessments if the mean
noninteger ranks that we assigned to stimuli correlation coefficient equaled or exceeded the
selected on the same number of trials, we critical r value of .58 (Salkind, 2001). Strong
PREFERENCE STABILITY 195
Figure 2. A description of preference stability for each participant is shown in each of the 10 panels. Mean rank
ordercorrelation coefficients werederivedbycomparingranks fromeachpreference assessment withranks fromevery
otherpreferenceassessment.Pointsabovethedottedlinesindicatestrongpositivecorrelations,exceedingacriticalr value.
positive correlations were exclusively observed Gina, Han, Inga, and Jan in Figure 2), in that
with Han (Figure 2), consistent with the the majority of mean correlation coefficients
bar-chart data that showed Han’s repeated met or exceeded our critical r value. Stability
assessment data to be quite stable (Figure 1). was difficult to determine for Dana, whose
By contrast, Abe’s mean correlation coefficients coefficientsvariedaboveandbelowthecriterion
were quite variable, with all falling below our throughout her evaluation.
critical r value, also consistent with his Table 1 shows that the mean correlations
graphically depicted data. between preference assessments for each partic-
Preference measures appeared to be highly ipant mapped onto the conclusions derived
unstable for 2 of the participants (Abe and Bo; from Figure 2 in that preferences were deemed
Figure 2). Consistently low mean correlation highly variable for 2 participants, stable for 7
coefficients showed that Bo’s assessment out- participants, and slightly below stability criteri-
comes were the most inconsistent, whereas with on for 1 participant. By comparing each
Abe, assessment outcomes were becoming more assessment to every other assessment, we were
consistent across repeated exposures. Preference able to determine the consistency of preference
measures appeared to be generally stable for 7 measures collected with the same participant;
participants (see the data for Cathy, Ed, Fay, hence, a measure of preference stability was
196 GREGORY P. HANLEY et al.
derived.Incontrast,bycomparingtheresultsof satiation and conditioning procedures were
the initial assessment with those of each removed from the radio (M rank 5 2.5) and
subsequent assessment, we were able to derive bubbles (M rank 5 9), respectively. Inga’s
ameasureofthepredictivevalidityoftheinitial preference rankings for the office task (second
preference assessment. panel) were variable in baseline but were
The mean of the correlations between the consistently high just prior to arranging daily
first preference assessment and each subsequent access to this activity (M rank 5 3.4). By
one are shown in Table 1 for each participant. contrast, the writing pad was consistently
These data are important because preference ranked low during the same baseline period
assessments are often conducted only once, and (M rank 5 7.4). When the satiation and
the capacity to predict preferences at a different conditioning procedures were applied to the
point in time based on the initial assessment office task and the writing pad, respectively, the
resultsisintegralforthesuccessof teachingand rank for the office task decreased (M rank 5
behavior-management programs. The correla- 5.7), whereas the rank for the writing pad
tion coefficients derived from comparing the increased immediately (M rank 5 4.2).
initial assessment with each subsequent assess- The third panel of Figure 3 shows the ranks
mentwerefoundtobeabovethecriticalrvalue for the two items that were present during
for half of the participants. conditioning sessions but were not paired with
ThebottompanelsofFigure 2showtherank social or edible reinforcers. The fourth panel
order correlation coefficients for Inga and Jan, shows the remaining three items that were
the 2 individuals who participated in the included in each preference assessment. Data
satiation and conditioning procedures. Their for these five items show that there were no
baselines were stable, in that all but one systematic changes in rank following imple-
assessment shared a strong positive correlation mentation of the initial or subsequent satiation
with the other assessments. Subsequently, and conditioning procedures for the other four
variability was imposed on their stable prefer- items.
ence patterns by providing free access to their Similar shifts in rank as a function of the
highly preferred items on days prior to the satiation and conditioning procedures were
preference assessment (the satiation procedure) observed with Jan (Figure 4, top two panels).
and by pairing originally nonpreferred items During baseline, the radio was ranked first
with social and consumable reinforcers (the during all but one assessment (M rank 5 1.7),
conditioning procedure). Figure 3 provides whereas drawing was consistently ranked low
a detailed examination of the independent (Mrank58.3).Gradualshiftsinrankforthese
contributions of the satiation and conditioning activitieswereobservedduringthesatiationand
procedures with Inga and Jan. The top panel conditioning conditions. Preference for the
shows that the radio was consistently ranked radio (satiation procedure) steadily decreased
first (M rank 5 1) and that the bubbles activity and was ranked 9 for the last two sessions,
was ranked low (M rank 5 8.3) in the majority whereas the rank of the drawing book (condi-
of Inga’s baseline assessments. Her preference tioning procedure) increased (M rank 5 7.4).
for the radio was disrupted during the satiation During the return to baseline, the rank of the
procedure,asevidencedbyadecreasingtrendin radio increased (M rank 5 4.8), although the
rank (M rank 5 6.2). By contrast, the rank for level observed in the initial baseline was not
the bubbles activity shows a slight but notice- recovered, and the rank of the drawing book
able increase (M rank 5 5.8). Baseline decreased to previously observed levels (M rank
preference ranks were recovered when the 5 8.8). The high rank associated with the
PREFERENCE STABILITY 197
Figure3. TheresultsoftheexperimentalanalysisofpreferencestabilityareshownforInga.Thefirstandsecondpanels
showtheresultsofthesatiationandconditioningproceduresforthefirstandsecondsetofitems,respectively.Thethird
panelshowstherankforthetwoitemsthatwerepresentduringconditioningsessionsbutwerenotpairedwithsocialor
ediblereinforcers.Thefourthpanelshowstheremainingthreeitemsthatwereincludedineachpreferenceassessment.
198 GREGORY P. HANLEY et al.
Figure 4. The results of the experimental analysis of preference stability are shown for Jan. The first and second
panelsshowtheresultsofthesatiationandconditioningproceduresforthefirstandsecondsetofitems,respectively.The
third panel shows the rank for the two items that were present during conditioning sessions but were not paired with
social or edible reinforcers. The fourth panel shows the remaining three items that were included in each
preference assessment.