Efficacy Trial of Contrasting Approaches to the Response-Contingent Learning of Young Children with Significant Developmental Delays and Multiple Disabilities

Findings from a randomized control design efficacy trial of an asset-based vs. needs-based approach to the response-contingent learning of infants and toddlers with significant developmental delays and disabilities who did not use instrumental behavior to produce or elicit reinforcing consequences are reported. The investigation included 71 children randomly assigned to the two contrasting types of interventions. The asset-based intervention and needs-based intervention differed in terms of how child behavior were identified and used to elicit reinforcing consequences as part of response-contingent learning games implemented by the children’s parents or other primary caregivers. Children in the asset-based group were provided more learning opportunities, acquired more response-contingent behavior, and demonstrated more efficient learning compared to children in the needs-based group. Implications for improving practices for very young children with significant developmental delays and multiple disabilities are described.


Introduction
Four decades ago, Maier (1978) defined intervention as "A series of socially engineered … activities in which a practitioner deliberately introduces specifically structured events into the experience of an individual or group of individuals … in order to facilitate ordinary developmental processes " (p. 195).A year later, Garwood (1979) defined early childhood intervention as the "provision of meaningful and appropriate educational experiences to preschool children to maximize educational benefits" (p.3).Whether stated explicitly or implicitly, all interventions are developed from different paradigms where different worldviews include different beliefs and presuppositions about the experiences hypothesized to be related to outcomes of interest (Reese & Overton, 1980).These beliefs and presuppositions in turn influence decisions about both research and practice (Hartung, 2015;Turner & Reese, 1980).Woodhead (2006) noted that historically different early childhood intervention paradigms have influenced policy and practice in discernibly different ways.

Intervention Paradigms
Two contrasting paradigms have most often been used to conceptualize and operationalize early intervention practices with infants and toddlers with developmental delays and disabilities.One approach focuses on identifying delayed or missing skills and then intervening to teach or facilitate children's acquisition of those skills (Allen & Cowdery, 2015;New & Cochran, 2007).The other approach uses existing child behavior as the building blocks for promoting acquisition of instrumental and functional behavior (Leffert, Benson, & Roehlkepartain, 1997;Wilson, Mott, & Batman, 2004).Eloff and Ebersohn (2001) described these contrasting approaches to early intervention as needs-based and asset-based approaches respectively.
The two approaches differ in terms of their assumptions about the purpose of intervention.Asset-based intervention is premised on the belief that human growth and development is best facilitated when existing behavioral competencies are changed or modified as a result of new or challenging learning opportunities.In contrast, needs-based intervention is premised on the belief that foundational behavior not in a person's repertoire need to be targeted and taught using practices intended to promote acquisition of those behavior.Asset-based approaches to intervention focus on building on child strengths, interests, preferences, and other skills to enhance learning (e.g., Dunst, Herter, & Shields, 2000;Eloff & de Wet, 2009;Lancioni, Singh, O'Reilly, Oliva, & Groeneweg, 2005), whereas needs-based approaches to intervention focus on the identification of child weaknesses, deficits, or missing skills and the amelioration of these deficiencies (see Dunst & Trivette, 2009;and Panitch, 1993; for descriptions of the features of this approach to early intervention).

Purpose of the Study
The purpose of the study described in this paper was to compare an asset-based versus needs-based approach to response-contingent learning of infants and toddlers with significant developmental delays and multiple disabilities to determine if the two approaches differed in terms of their consequences operationally defined in terms of a number of different child learning measures.Response-contingent learning refers to a child's ability to use a behavior to produce or elicit reinforcing environmental consequences (Gunnar, 1980;Williams, 2001).This type of instrumental learning was described by Watson and Ramey (1972) as response-contingent stimulation because an environmental effect (stimulation) is contingent upon a child's behavioral response, whereas Piaget (1952) described this type of learning as secondary circular reactions where a child's repeated actions on the social and nonsocial environment result in interesting consequences.According to both Gunnar (1980) and Tarabulsy et al. (1996), this type of learning is important because contingency experiences provide infants opportunities to learn to control environmental events and to recognize that they are the agents of environmental effects (i.e., contingency awareness).

Previous Research
More than 60 years of research indicates that infants without delays or disabilities as young as 2 to 4 months of age acquire the ability to use vocalizations, smiling, visual attention, arm and leg movements, head turns, sucking movements, and other behavior to produce or elicit environmental consequences (see Dunst, Gorman, & Hamby, 2010;Dunst, Storck, Hutto, & Snyder, 2007;Hulsebus, 1973;Lipsitt, 1969;Rovee-Collier & Gekoski, 1979; for reviews).Infants and toddlers with disabilities or delays often take longer to learn these same types of response-contingent behavior and as a result often demonstrate a latency to learn (Hutto, 2007).The extent to which a latency to learn response-contingent behavior among infants and toddlers with delays or disabilities is related to different types of experiences or interventions has not been directly examined nor have the differences in rates of learning been explicitly determined (see e.g., Dunst, Gorman et al., 2010;Dunst, Storck et al., 2007).
Close inspection of response-contingent learning studies with infants and toddlers with significant delays and multiple disabilities suggest that the characteristics of different interventions have differential consequences in terms of child learning.For example, Dunst, Cushing, and Vance (1985) used a needs-based approach to contingency learning where behavior that children did not yet exhibit were targeted to be learned and found a latency to learn for half of the children in their study.In contrast, Lancioni et al. (2005) used an existing child behavior and favorite stimuli to facilitate a young girl's use of vocalizations to produce a preferred environmental consequence.Dunst et al. (2007) reported similar results using an asset-based approach to response-contingent child learning with young children with delays and disabilities.The differences between the two approaches to child learning were the basis for hypothesizing that an asset-based intervention would be more effective than a needs-based intervention for promoting the response-contingent learning of young children with significant delays and multiple disabilities.

Study Hypotheses
The study described in this paper was a randomized controlled efficacy trial that included different measures of child learning opportunities, child response-contingent learning, and child learning efficiency.The four hypotheses of the study were: H 1 : Children in the asset-based group will be afforded more learning opportunities (number of learning trials per game) compared to the children in the needs-based group.
H 2 : Children in the asset-based group will demonstrate acquisition of more response-contingent behavior (total number of contingency responses and average number of contingency responses per learning game) compared to the children in the needs-based group.
H 3 : The response-contingent learning opportunities afforded the children in the asset-based group will result in more efficient learning (percent of trials associated with response-contingent behavior and average number of response-contingent behavior per minute) compared to the children in the needs-based group.
H 4 : The rates of change on the different child learning measures will increase more rapidly among the children in the asset-based group compared to those for the children in the needs-based group.
The study is part of a line of research and practice investigating the use of response-contingent learning opportunities with children with significant developmental delays or multiple disabilities to enhance their abilities to interact with their social and nonsocial environments in ways having reinforcing consequences (e.g., Dunst et al., 1985;Dunst, Raab, Trivette, Parkey et al., 2007;Dunst, Raab et al., 2010;Dunst, Raab, Wilson et al., 2007;Raab, Dunst, Wilson, & Parkey, 2009).Preliminary analyses of data from the study yielded encouraging findings about the differences in the two approaches to response-contingent learning (Raab, Dunst, & Hamby, 2016).The study described in this paper included a larger sample of children and different types of data analysis procedures to evaluate whether the two approaches to response-contingent learning had differential effects.Hierarchical linear growth curve modeling was used to examine intraindividual changes in the children's learning and the extent to which those changes were differentially related to the contrasting types of interventions.The study results were also examined in terms of the relative effectiveness of the asset-based intervention beyond that associated with the needs-based intervention.Findings were expected to either support or refute the contention that the two approaches to response-contingent learning had different characteristics and consequences.

Institutional Review and Approval
The study described in this paper was reviewed and approved by the investigators' Research Institute Institutional Review Board.This included, but was not limited to, the approval of the intervention procedures, the procedures for obtaining participant informed consent, and the procedural safeguards for protecting the study participants.

Recruitment
The children were recruited from early intervention programs, preschool special education programs, hospitals, specialty clinics, physician practices, and parent and disability organizations in three states in the southeastern United States.Children were eligible for participation in the study if they had multiple disabilities (e.g., cerebral palsy and sensory impairments), identified disabilities associated with significant developmental delays (e.g., congenital anomalies or genetic disorders), significant developmental delays without known causes, neurologic diseases or central nervous system disorders (e.g., lissencephaly, Ohtahara syndrome), or birth-related conditions associated with poor developmental outcomes (e.g., extreme low birth weight and grade 3 or 4 intraventricular hemorrhaging); were functioning at or below a 4 to 5 month age level of development; and not yet demonstrating intentional use of behavior to produce or elicit reinforcing social or nonsocial environmental consequences.
One hundred and twenty children were referred to the study and screened for eligibility.Eighty-one children were determined to be eligible for participation.Children who met the eligibility criteria were randomly assigned to either the asset-based (N = 42) or needs-based (N = 39) intervention groups.Four children were lost to attrition in the asset-based group (10%) and six children were lost to attrition in the needs-based group (15%).Two parents in each group never gave oral or written consent to participate, and the investigators lost contact with the other six families.Both overall attrition (10%) and differential attrition (5%) were small enough to establish a low level of bias (What Works Clearinghouse, 2014, n.d.-a).The final sample sizes were 38 and 33 respectively in the asset-based and needs-based intervention groups.

Participants
The participants were 71 children (37 boys and 34 girls) birth to six years of age and their parents (65 mothers, 2 fathers) and other primary caregivers (2 foster parents, 2 parent guardians).There were no differences in the proportion of males and females in the two intervention groups, χ 2 = 0.74, df = 1, p = .390,d = .21,nor was there a difference in the proportion of mothers in the two intervention groups, χ 2 = 1.39, df = 1, p = .239,d = .28.These findings indicate that the two samples were more similar than different on those participant characteristics.
The Mullen (1995) Scales of Early Learning were administered to each child at entry into the study.The children were, on average, 17 months of age (SD = 10) but functioning developmentally, on average, at only 4 months of age (SD = 3).Fifty-six percent of the children had Mullen Composite Standard Scores of 49 which is the lowest score attainable.A score of 49 or less is three or more standard deviations below a mean of 100 where 90 percent of the children had standard scores that indicated very low functioning levels (Mullen, 1995).
Because the majority of the children had Mullen Composite Standard Scores at or below 49, the children's Developmental Quotients (DQ) were estimated by dividing each child's composite developmental age on the Mullen scales by his or her chronological age and multiplying the result by 100.The children's average DQ was 34 (SD = 25).Seventy-five percent of the children had DQs three or more standard deviations below a mean of 100, 60% had DQs four or more standard deviations below the mean, and 50% of the children had DQs five or more standard deviations below the mean.
Table 1 shows the developmental characteristics of the children and the background characteristics of the parents in the two intervention groups at entry into the study.The children's chronological ages, developmental ages, and DQs were very much alike in both groups as evidenced by no statistically significant between group differences.There were also no statistically significant between group differences in the parents' ages, years of formal education, or family socioeconomic status (Hollingshead, 1975).The results are an indication that randomization resulted in the children and parents in the two intervention groups being more similar than different.There was, however, nonbaseline equivalence for child DQs and parent education as evidenced by mean difference effect sizes of d = .31and d = .36,respectively (What Works Clearinghouse, 2014, n.d.-b).Both variables were used as covariates in all of the analyses described below.Inasmuch as the study participants were also receiving special instruction/special education, and therapy services from other programs, agencies, and providers, we ascertained the similarity of the two groups of children on early childhood intervention services.The percent of children receiving the four services that "make up" the bulk of early childhood intervention in the U.S. (Hebbeler et al., 2007) are shown in Table 2.There were no between group differences for any of the four comparisons.There were also no differences in the total number of the four kinds of services for the children in the asset-based (M = 2.55, SD = 1.41) vs. needs-based (M = 2.39, SD = 1.09) intervention groups, t = 0.54, df = 69, p = .593,d = .13.

Procedure
Three early childhood practitioners were assigned to implement the asset-based intervention and three early childhood practitioners were assigned to implement the needs-based intervention.The asset-based approach used existing behavior in the children's repertoire as target behavior as part of response-contingent learning opportunities, whereas the needs-based approach used adult-selected behavior for the children to learn as target behavior.The practitioners in the two groups were kept unaware of the fact that there was a contrasting intervention group, and the staff assigned to the two different intervention groups never interacted with one another throughout the course of the study.
The children in both groups were first assessed to identify behavior that the children would use to produce or elicit environmental consequences as part of response-contingent learning opportunities (described to the parents as learning games).The two interventions differed only in the procedures used to identify and select target behavior.Lancioni et al. (2001) described the differences between the contrasting interventions in terms of asset-based practices "not requiring excessive effort" (p.271) to control environmental consequences and needs-based practices as requiring "excessively high levels of effort" (p.271) to control environmental consequences.
The children in the asset-based group were observed in everyday activities in their homes and their parents queried to identify behavior in the children's repertoires but not used intentionally to produce or elicit reinforcing environmental effects.An investigator-developed checklist was used to record the occurrence and frequency of different child behavior, including, but not limited to, head, body, arm, leg, fist, and hand movements; vocalizations; and directed gaze and visual fixation.Behavior that a child produced frequently or for considerable durations of time were selected as target behavior to be used as part of providing the children response-contingent learning opportunities.
The children in the needs-based group were administered the birth to three-year-old Assessment, Evaluation, and Programming Systems (AEPS) (Bricker, Capt, & Pretti-Frontczak, 2002) scales to identify missing skills operationalized as behavior at or just above the ceiling level in six scale domains (fine motor, gross motor, adaptive, cognitive, social-communication, and social).Results were used to select child behavior as the intervention targets where response-contingent learning games were used to facilitate acquisition of those behavior targets.The AEPS was not used as intended by assessment system developers but rather to simply identify intervention targets for response-contingent learning opportunities.
As part of the assessments administered to both groups of children, child-specific reinforcers were identified through observations and parent interviews to determine the people, materials, activities, events, and so forth that elicited or maintained the children's attention.This was done in a manner similar to how Dunst, Raab, Trivette, Wilson et al. (2007) identified reinforcing consequences in their research with children with significant developmental delays and multiple disabilities.Fisher, Piazza, Bowman, and Amari (1996) noted that this is a more effective approach to identify reinforcers compared to using a predetermined list of potential reinforcers.
The same types of response-contingent learning games for children in both intervention groups were used to promote the children's use of targeted behavior to produce or elicit reinforcing consequences.Learning games included targeted operant behavior that either resulted in reinforcing consequences (e.g., swiping at a mobile producing movement or sound) or were reinforced by a caregiver (e.g., an adult talking to a child each time he or she looked at the adult's face).All of the learning games were characterized by behavior-based contingencies where the availability of a reinforcement or the production of an interesting consequence was dependent on the children's production of the targeted behavior (Tarabulsy et al., 1996).

Implementation
Early childhood staff were taught the intervention practices in the contrasting approaches over a two-month period of time prior to their work with children and their families using an evidence-based adult learning procedure (Dunst, Trivette, & Hamby, 2010).These early childhood staff, in turn, taught parents to use the practices in the families' homes using a modified version of the adult learning procedures (Raab, Dunst, & Trivette, 2013).
The adult learning procedure involved staff's active participation in four phases for learning either of the two interventions: (1) acquiring information about and examples of the intervention practices; (2) authentic use of the interventions and evaluating the characteristics and outcomes of the practices; (3) reflecting on their overall understanding and mastery of the practices; and (4) identifying and participating in additional opportunities to learn to use their particular approach to intervention.Staff were trained using procedural manuals that were identical for the two intervention groups except for the methods used to identify target behaviors.Video examples of response-contingent learning opportunities were used both to illustrate the practices to the practitioners and to provide staff opportunities to use and understand how the interventions were implemented.Role-playing, in-vivo training, and investigator feedback were employed to provide staff opportunities to use the interventions in a manner consistent with the study procedures.
The practitioners and the children's parents developed four to six learning games so that targeted behavior were associated with child-specific reinforcers to facilitate the children's acquisition of contingency responses.Both social and nonsocial games were developed for each child.Social learning games involved a child's use of a target behavior to elicit a social response from the parent (e.g., a parent kissing the child on the neck each time she turned her head from side to midline while in a supine position).Nonsocial learning games involved a child's use of a target behavior to elicit a nonsocial environmental consequence (e.g., a child producing movement of a mobile by leg or hand movements).
Parents were taught to implement the learning games by the practitioners using the same adult learning procedure used to train the staff (Dunst et al., 2010).For each game that was developed, staff engaged a parent in completing an investigator-developed Learning Games Intervention Form that included the target behavior and reinforcing consequence used in a game, procedures to prepare and set up a game (e.g., materials needed, positioning of the child, etc.), strategies for implementing the game to ensure delivery of the reinforcing consequence when the child produced a target behavior, and any special child-specific considerations (e.g., wait time, adaptations, etc.).Staff described the key characteristics of a game, demonstrated the use of the game, and then had the parent practice each game with her child.Supportive feedback and guidance were provided as necessary to ensure the parents understood how to implement the learning games in the ways intended.
The practitioners visited the children and parents weekly or every other week and observed and recorded the types of games played with the children, the length of time each game was played, the child behavior that was targeted to elicit or produce reinforcing consequences for each game, and the number of child behavior that had and did not have reinforcing consequences.The home visits were also used to change or modify the learning games, develop new learning games, and to coach the parents in how to engage their children in the learning games.

Procedural Fidelity
The parents maintained weekly logs, which were used to determine procedural fidelity in terms of the number of games developed for a child compared to the number of games implemented by a parent with the child, and the frequency of child participation in the games per week.There were, on average, 2.92 (SD = .79)different games planned and 2.72 (SD = .87)games implemented for each child per week, t = .61,df = 69, p = .545,d = .15.The games were played, on average, 4.30 days per week (SD = 1.38).There was no statistically significant difference in the number of days per week that the children participated in the games in the two intervention groups, t = .08,df = 69, p = .933,d = .02.These results indicated that the groups did not differ in terms of the number of planned compared to the number of implemented games or how often the parents implemented the games with the children in the two intervention groups.

Dependent Measures
Investigator-developed recording forms were used to code different information about the learning games during the home visits by the staff with the children and their parents.Each recording form included space for recording the number of trials per game.A trial was defined as the availability of a child-specific reinforcement where a targeted behavior for a game (either prompted or non-prompted) did or did not result in the reinforcing consequence.The recording form included codes for non-prompted targeted behavior that produced or elicited a reinforcing consequence, physically and verbally prompted child contingency behavior, and child behavior that had no environmental effects per learning opportunity trial.A trial lasted up to six seconds at which time the child's responses were coded.A learning game was defined as a series of learning trials that involved use of a targeted behavior to produce or elicit a child-specific reinforcing consequence where a game lasted until a child ceased attempts to elicit a reinforcing consequence, or lost interest in a game.Each child had an average of five data collection sessions during the eight weeks of intervention.Three games on average were played with each child per session where each game lasted an average of six minutes.
The information on the recording forms was used to compute two child learning opportunity measures, two child response-contingent behavior measures, and two child learning efficiency measures.The child learning opportunity measures included the number of learning games used with a child and the number of learning opportunities (trials) per game.The child response-contingent behavior measures included the total number of non-prompted response-contingent behavior that resulted in a reinforcing consequence and the average number of non-prompted child response-contingent behavior per learning game.A non-prompted response-contingent behavior was defined as a child's use of a targeted behavior to produce or elicit a reinforcing consequence that was not prompted by either a parents' or a practitioners' verbal or nonverbal behavior.The child learning efficiency measures included the percent of learning opportunities (trials) associated with non-prompted child behavior having reinforcing consequences and the average number of non-prompted child behavior having reinforcing consequences per game minute.The six learning measures were first computed for each game and then averaged across games per session.

Interrater Agreement
The early childhood practitioners made a total of 355 home visits during the course of the study where 1,010 learning games were played during the home visits.Research assistants (one for each intervention group) made joint visits with the practitioners in their respective intervention groups on 95 occasions (27% of the total number of home visits) where 272 learning games (27% of the total number of learning games) were played during the home visits.The research assistants used the same recording form as the practitioners where the two sets of codes were used to compute interrater agreement for (1) the number of games played during the home visits, (2) the length of the games in minutes, (3) the number of learning trials per game, and (4) the number of non-prompted response-contingent child behavior.Interrater agreement was computed as the number of agreements divided by the number of agreements plus non-agreements multiplied by 100.There was 100% agreement on both the number of games played during the home visits and the number of minutes each game was played.There was 92% agreement on the number of trials for the learning games (range = 89 to 93), and 91% agreement on the number of non-prompted child response-contingent behavior (range = 85 to 95).

Data Analysis
Hierarchical linear growth curve modeling (Raudenbush, Bryk, Cheong, Congdon, & du Toit, 2011) was used to evaluate the effects of the two contrasting approaches to interventions on the six child learning measures.Each analysis included tests for linear increases in the slopes of the learning measures (Level-1) and between intervention group differences in the slopes controlling for the non-time varying covariates (Level-2).The between group comparisons are analogous to between group x measurement occasion tests for interactions between the independent and dependent measures.The growth curves were modeled with each child's baseline learning measure equal to zero inasmuch as none of the children demonstrated instrumental response-contingent behavior at the beginning of the study.The main analyses were supplemented by additional analyses as indicated to clarify the nature of the relationships between the two contrasting interventions and the child learning measures.
In addition to statistical significance testing, we also computed Cohen's d effect sizes for the differences in the slopes for the two intervention groups.The effect sizes were computed using the formula d = b/√tau, where b is the regression coefficient for the between slope differences and tau is the random effects variance component for the measurement occasion slopes (Feingold, 2009).We also computed the improvement indices for the between group slope differences for evaluating the practical importance of the intervention effects (What Works Clearinghouse, 2014).This is a measure of the relative effectiveness of the asset-based intervention beyond that associated with the needs-based intervention (Lipsey et al., 2012).An improvement index can take on values between -50 and +50, where positive indices show favorable results for the particular intervention hypothesized to be more effective.It is interpreted as the percentage of slope indices in the asset-based group that exceeded those in the needs-based group.ZCalc was used to compute the improvement indices (Neill, 2006).

Results
Table 3 shows the results for the linear growth curve analyses for each of the six dependent measures with the effects of the non-time varying covariates (parent education and child DQ) partialled from the between group differences.Parent education was not related to any of the between group slope differences, and child DQ was related to only the between group slope differences for the total number of nonprompted response-contingent behavior, t = 2.04, df = 67, p = .045,d = .50,and average number of response-contingent behavior per game, t = 2.22, df = 67, p = .030,d = .54.
Results indicated that there were statistically significant between group slope differences in the rates of change for 4 of the 6 learning measures, but large to very large effect sizes for the between group slope differences for all six learning measures.All of the sizes of effect favored the asset-based group except for the number of learning games played with the children.The findings showed that the progressive increases in the slopes on the dependent measures (except the number of learning games) were greater for the asset-based intervention compared to the needs-based intervention.Notes.Results are adjusted for baseline differences in child DQs and parent education.d = Cohen's d effect size.

Child Learning Opportunities
There were linear increases in the number of learning games played with the children across measurement occasions for all the children taken together, b = .39,t = 20.46,p = .000,d = 4.76, where the linear growth curve slopes were greater for the needs-based intervention compared to the asset-based intervention (Table 3).There were also linear increases in the number of learning opportunities (trials) afforded the children, b = 1.54, t = 9.101, p = .000,d = 1.56,where the linear increases were much more pronounced for the asset-based compared to the needs-based intervention.
The average slopes for both child learning opportunity measures are shown in Table 4.The average slope for the number of learning games was b = .46for the needs-based group compared to b = .33for the asset-based group.This finding, however, is somewhat misleading because of the small standard error associated with the between slope differences.This is the case as shown in Figure 1 where the slopes for the two contrasting interventions are very much alike.The average number of games played at Week 1 was 1.67 (SD = .15)for the asset-based group and 1.46 (SD = .11)for the needs-based group, and at Week 8 was 3.78 (SD = .32)for the asset-based group and 3.91 (SD = .26)for the needs-based group.
Although the number of learning games played per week with the children in both intervention groups were more similar than different, the rates of increase in the number of learning opportunities (trials) afforded the children per game in the asset-based group (b = 1.94) was nearly twice as large as that for the needs-based group (b = 1.09).The between slope difference shown in Table 4 was associated with a very large effect size (d = 1.13).
The difference in the slopes is shown in Figure 2 where the average number of trials per game at the end of the study was 20.96 (SD = 2.41) for the asset-based group compared to 10.63 (SD = 1.75) for the needs-based group, t = 20.37,df = 69, p = .000,d = 4.90.The between group difference in the slopes for number of child learning opportunities had an improvement index of 37.This indicates that if the children in the needs-based group had received the asset-based intervention they would have experienced 37% more learning opportunities (trials) per game.Notes.The average slopes are adjusted for the two non-time varying covariates (child DQ and parent education).The Cohen's d effect sizes for the Table 3 contrasts were used to compute the improvement indices (Neill, 2006).
Figure 1.Average number of learning games implemented during each week of intervention

Child Response-Contingent Behavior
The between intervention group linear trend comparisons for the total number of non-prompted response-contingent behavior and average number of non-prompted response-contingent behavior per game were both associated with statistically significant differences and very large effect sizes (Table 3).In both analyses, the asset-based group demonstrated faster rates of learning compared to the needs-based group (Table 4).The rates of change for both the total number of response-contingent behavior and the average number of response-contingent behavior per game were more than three times greater for the asset-based intervention compared to the needs-based intervention.Both differences were associated with very large between slope differences effect sizes.

Child Learning Efficiency
Results showed that the between group differences in the slopes for the percent of response-contingent behavior per number of trials was associated with a large effect size (d = .86)and the differences in the average number of response-contingent behavior per minute was also associated with a very large effect size (d = 4.87).Both sets of analyses showed that learning efficiency improved more rapidly for the asset-based intervention group compared to the needs-based intervention group (Table 4).At the end of the eight weeks of the intervention, the average percent of trials associated with nonprompted response-contingent behavior was 99.99 (SD = 6.42) for the asset-based intervention compared to 52.36 (SD = 4.78) for the needs-based intervention, t = 35.03,df = 69, p = .000,d = 8.42.
Closer inspection of the findings showed a curvilinear rather than a linear relationship for the changes in the percent of trials associated with nonprompted response-contingent behavior and a between slope difference in the rates of change between the contrasting interventions, b = 2.02, t = 6.85, p = .000,d = 4.12.Figure 4 shows the differences in the curvilinear rates of change for the two contrasting interventions.As hypothesized, child response-contingent learning was considerably more efficient for the asset-based intervention compared to the needs-based intervention as evidenced by a more rapid increase in the percent of trials associated with response-contingent behavior during the first three weeks of intervention and a larger and more stable ratio of response-contingent behavior to learning trials during the last three weeks of intervention.
Both learning efficiency measures had improvement indices favoring the asset-based intervention (Table 4).These indices show the relative efficiency of asset-based contingency learning practices beyond those associated with needs-based contingency learning practices.The results, and especially those for the percent of response-contingent behavior per number of learning trials (Figure 4), illustrate the value-added effects of an asset-based approach to early intervention.It is important to note that the effects of the asset-based intervention were considerably more robust than found in earlier studies with children with significant developmental delays or disabilities (e.g., Bailey & Meyerson, 1969;Haskett & Hollar, 1978;O'Brien, Glenn, & Cunningham, 1994;Utley, Duncan, Strain, & Scanlon, 1983) including earlier studies conducted as part of the line of research and practice by the study investigators (e.g., Dunst et al., 1985;Dunst, Raab, Trivette, Parkey et al., 2007;Dunst, Raab et al., 2010;Dunst, Raab, Trivette, Wilson et al., 2007;Raab et al., 2009).We also note that the acquisition and use of response-contingent behavior in the asset-based group was almost immediate and there was no latency to learn as has been the case in other studies of young children with severe developmental delays and multiple disabilities (Hutto, 2007).These findings are most likely the case because the asset-based intervention was conceptualized and implemented from a strengths-based perspective where the children were seen as having existing capabilities as well as the capacity to become more competent.This included the use of existing child behavior as the building blocks for developing their response-contingent capabilities and child-specific reinforcers which was not necessarily the case in previous research where targeted behavior and reinforcers were predominantly investigator or practitioner selected rather than based on existing behavior in a child's repertoire (e.g., Hanson & Hanline, 1985;Sullivan, Laverick, & Lewis, 1995;Watson, Hayes, & Vietze, 1982) including our own research (e.g., Dunst et al., 1985;Laub & Dunst, 1974).
The fact that the children in the needs-based group made some progress deserves comment because the intervention practice used with this group of children was intended to mirror a particular type of early childhood intervention where the results from developmental assessment scales are used to identify behavior for children to learn.This approach to intervention is part of a long standing tradition of using assessment scale results as the source of information for identifying behavior that early childhood intervention is designed to promote or facilitate but which children do not yet use in an intentional or functional manner (see especially Dunst, Snyder, & Mankinen, 1988).As stated by Eloff and Ebersöhn (2001), "Even though many proponents of the needs-based approach ... [include assessment] of strengths and weaknesses, practice serves to emphasize weaknesses" (p.149).
The learning demonstrated by the children in the needs-based group is typical for the type of intervention that was used with the children in the study described in this paper (Hutto, 2007).In fact, infants and toddlers with significant developmental delays and young children with multiple disabilities often demonstrate at least some learning as a result of any number of early childhood intervention practices (see e.g., Dunst, 1986).In the absence of the findings from the children in the asset-based group, one might conclude that the needs-based intervention was about as effective as would be expected with children with significant delays and disabilities.
The results show that the needs-based intervention was not nearly as effective as the asset-based intervention as evidenced by the results reported in the paper.
The practical significance of the asset-based approach to response-contingent learning was demonstrated by the magnitudes of the improvement indices for 5 of the 6 learning measures (Table 4).The magnitudes of the improvement indices are larger than those reported for other types of early childhood interventions (e.g., Feil, Frey, Walker, & Forness, 2015;What Works Clearinghouse, 2006), and illustrate the value-added effects of asset-based compared to needs-based approaches to response-contingent intervention.
The differences between asset-based and needs-based practices perhaps is best illustrated by noting that an asset-based approach is part of a family of early childhood intervention practices that emphasize child and family strengths as the behavior used to promote participation in everyday activities to provide opportunities to interact with people and material in ways that not only strengthen existing capabilities but also provide contexts for acquiring new skills (e.g., Campbell, Milbourne, & Silverman, 2001;Granlund, Wilder, & Almqvist, 2013;Green, McAllister, & Tarte, 2004;Swanson, Raab, & Dunst, 2011).Accordingly, the learning games for the asset-based group were the sources of child learning opportunities, and existing non-instrumental child behavior were the strengths (abilities) used to elicit reinforcing environmental consequences and promote child acquisition of instrumental and functional behavioral capabilities as was found in the study described in this paper.

Implications for Practice
We conclude by noting the implications for changing and improving early childhood intervention practices.Young children with significant delays and disabilities more often than not are seen as lacking foundational skills that lead to identification of child behavior, which the children do not or cannot easily produce (i.e., missing skills).Results reported in this paper and elsewhere (e.g., Dunst, Raab, Wilson et al., 2007;O'Brien et al., 1994) indicate that intervention practices for young children with significant developmental delays or multiple disabilities could be improved considerably by (1) changing the ways in which target behavior are selected, (2) using assessment procedures for identifying child-specific reinforcers, and (3) arranging learning opportunities (e.g., learning games) to facilitate child use of behavior to produce or elicit reinforcing environmental consequences (see e.g., Dunst, 1981;Lancioni, 1980;Sullivan & Lewis, 1990;Watson et al., 1982; for descriptions of response-contingent learning interventions).
An asset-based approach to child response-contingent learning starts by identifying behavior that children already demonstrate (intentionally or non-intentionally) and using these behavior as the building blocks to promote acquisition of contingency behavior (Eloff & Ebersöhn, 2001;Wilson et al., 2004).
Response-contingent learning is more likely to occur if child-specific rather than a priori selected reinforcers are used as the consequence of child behavioral responses (Crawford & Schuster, 1993;Piazza, Fisher, Hagopian, Bowman, & Toole, 1996).The experiences (e.g., learning games) afforded children to use behavior to produce or elicit reinforcing consequences need to be ones that have a high probability of occurring as frequently and as often as possible to provide lots of opportunities to be associated with positive effects (Dunst, Raab, Trivette, Wilson et al., 2007;Dunst, Trivette, Raab, & Masiello, 2008).This is the case, in part, because young children with delays or disabilities often take longer to learn contingency detection (Tarabulsy et al., 1996) and awareness (Watson, 1966), and the more children are afforded learning opportunities with developmentally enhancing characteristics, the more likely they will learn that they are the agents of environmental effects.

Figure 2 .
Figure 2. Average number of trials (learning opportunities) afforded the children per game across the eight weeks of intervention

Figure 3 .
Figure 3. Average number of child response-contingent behavior per game across the eight weeks of intervention

Figure 4 .
Figure 4. Average percentage of child response-contingent behavior per number of learning trials across the eight weeks of intervention

Table 1 .
Characteristics of the children and parents at entry into the study Hollingshead (1975)evelopmental age divided by child chronological age x 100.bHollingshead (1975)four factor index of socioeconomic status.

Table 2 .
Percent of children receiving early childhood intervention services

Table 3 .
Fixed effects estimates for the differences in the linear growth curves (slopes) for the asset-based vs. needs-based intervention groups

Table 4 .
Linear growth curve average rates of change for the contrasting types of interventions