Pages Navigation Menu

Generalization and Maintenance in Social Skills Research (1991)

Generalization and Maintenance in Social Skills Research (1991)

©1991, 2013 by Dallas Denny

Source: Generalization and maintenance in social skills research: An examination of two papers. Paper for Genderal Qualifying Examinations, Department of Special Education, George Peabody College of Vanderbilt University.




Generalization and Maintenance in Social Skills Research:

An Examination of Two Papers

By Dallas Denny

For General Qualifying Examinations

Department of Special Education

George Peabody College of Vanderbilt University

7 November, 1991



It is generally agreed that it is important to address issues of generalization and maintenance in social skills research. In this paper, I have examined two representative studies, looking specifically at the treatment of issues of generalization and maintenance by the authors.


The question of generalization and maintenance in applied behavior analysis is considered by most researchers to be one of crucial importance. Although Skinner addressed the issues of generalization (transfer of stimulus control to objects without a history of reinforcing the organism) and maintenance (exhibition of operant response after the discontinuation of the artificial reinforcement schedule) as early as 1938 in his work with nonhuman animals, it was Stokes & Baer (1977) who clearly and concisely brought the importance of these phenomenon to the attention of the behavior analytic community. Today, the importance of maintenance and generalization is widely discussed, but a number of researchers have lamented the fact that, like the weather, many researchers talk about these phenomena, but few adequately address (i.e. do much more than talk about) them in their studies. “Despite this longstanding concern with generalization and increasing attention to its assessment, it is difficult to point to any startling advances in our ability to reliably produce either generalization or maintenance of social behavior…” (Fox & McEvoy, manuscript in preparation).

The operant training of social skills is an area in which generalization and maintenance are of particular importance. This is because skills are typically trained in one or at most a few settings, are expected or hoped to transfer to new settings, and are hoped to endure after the experimenters have withdrawn, with the ever elusive “natural community of reinforcement” maintaining the shaped behavior across both time and place. Resources are scarce, and the need for permanent change is great. The purpose of this paper is to look at two representative articles from peer-refereed journals, examining their treatment of maintenance and generalization.


Fox, et al. (1984)

Fox, Gunter, Brady, Bambara, Spiegel-McGill, and Shores (1984) used multiple peer exemplars to develop generalized social responding in Linda, a 14-year-old girl with autism. Linda rarely interacted with other children. She was nonverbal, with some signing. Four age-matched nonhandicapped peers were trained to respond to Linda’s social bids, but not to initiate social interactions. The setting was a playroom in a public elementary school. During baseline, Linda and a peer were seated on the floor within reach of toys or games. Sessions began with a teacher saying, “It’s time to play.”

During the intervention phase, the teacher prompted Linda to make social initiations to the peer confederate; a graduated guidance procedure was used. Peer confederates had the same instructions as during baseline. The intervention, then, was prompting by the teacher.

Generalization sessions occurred after lunch, with Linda, her handicapped peers, and the four nonhandicapped peer confederates simultaneously entering a play area and being told, “It’s time to play.” Peer confederates had the same instructions as they did during the baseline and intervention phases, and behavioral observations were coded similarly.

An observational system used by Strain & Timm (1974) was used to measure interactions. Behaviors were coded as either “initiated, responded, or interaction events” (Fox, et al., 1984, p. 19). Behaviors which occurred more than three seconds after the social behavior of the other child were coded as initiated. Behaviors emitted within three seconds were classified as responded. If social interaction continued (within three seconds), it was termed an interaction.

Data were analyzed with the help of a microcomputer. Interobserver agreement was measured in both treatment and generalization settings.

A multiple baseline design (Hersen & Barlow, 1976) across peer confederates was used, with intervention phase begun at different times for the four nonhandicapped peer confederates. Daily, Linda was placed in the classroom with each peer confederate. Intervention phase was “switched on” for each peer confederate at a different time. The major independent variables were the frequency and time of social interactions.

Linda interacted only infrequently with the peer confederates during the baseline periods, with one exception, which I will discuss later. For the first three peer confederates, entering the intervention phase resulted in an increase in initiations by both Linda and her nonhandicapped peers, and in an increase in the time Linda spent in social interaction. Initiations to the fourth peer confederate increased dramatically shortly after the initiation of the intervention phase for the third confederate, while the fourth confederate was still in baseline.

After the 13th session (after initiation of the intervention phase for the second nonhandicapped peer confederate), Linda began to show increased frequency of initiations and an increase of time spent in interaction with non-handicapped peer confederates in the generalization setting. Both Linda and the peer confederates initiated these interactions. Rate of interaction with handicapped peers, who uniformly showed a low level of responding, did not change.

These results showed generalization across both nonhandicapped peer confederates and across settings. Unfortunately, Fox, et al. were not able to replicate, using another handicapped subject, and did not attempt to determine whether the effects maintained across time. However, further studies have shown the effect to be a real one.


Odom, et al. (1985)

Odom, Hoyson, Jamieson, and Strain (1985) used nonhandicapped peers to increase the rate of social interactions in three handicapped preschool children. The handicapping condition was not specified (labeled), but descriptions of the children suggest mental retardation with autistic-like behaviors. The study was conducted in three settings in a mainstreamed preschool classroom: during structured play; at a table, in which peers and subjects were seated next to each other; and in “learning centers,” in which peer confederates and students participated in a single activity.

Using adults as stand-ins for the subjects, nonhandicapped peers were trained to initiate desired social interactions, and to persist when their social bids did not result in a response. During intervention, peer confederates were rewarded with social reinforcers in the classroom session when their social initiations were responded to by the subjects.

The observational system used recognized six positive and two negative categories of social interactions. Data were kept during five minute sessions daily in each of the three settings, using 10-second interval recording. Interobserver agreement checks occurred across students, across settings, and across treatment conditions.

The experimental design was a multiple baseline (Hersen & Barlow, 1976) aross settings, replicated across the three subjects. After a baseline, intervention began in the structured play setting, and was later begun in the other two settings. During intervention, nonhandicapped peer confederates were rewarded for the subjects’ responses to their social bids by the teacher, who showed them “happy faces” as they earned them. Peer confederates were later reinforced by allowing them to count the number of happy faces they had earned. When peer confederates had not initiated for about 15-20 seconds, they were verbally prompted by the teacher. The major independent variable was peer-confederate initiations.

After the intervention had been “turned on” in all three settings, the reinforcement of the nonhandicapped peer confederates was withdrawn; they were told that this would happen, but that they should continue to initiate to the subjects. This, and the subsequent two phases, were started simultaneously in all three settings. A fourth phase resulted in a reduction of teacher prompts; they occurred only after the peer confederate had not initiated for one minute. A fifth phase resulted in the reinstatement of teacher prompts to previous levels.

For all three nonhandicapped peers confederates, social initiations were low during baseline in all three settings. Intervention resulted in an increase in positive initiations for all three peers and in all three settings, as treatment was “switched on.” Withdrawal of reinforcement did not have a major effect on any of the peer confederates in any of the conditions. However, reduction of teacher prompts resulted in a reduction of positive social initiations in all settings for all three peers confederates; the last phase (reintroduction of teacher prompts) resulted in a return to previous levels.

Results were similar for social interactions of the three students.

Odom, et al.’s peer-initiation treatment package effectively increased desirable social behaviors in all three handicapped subjects. Generalization did not occur across settings, however, and it was demonstrated that continued teacher prompting was necessary in order to maintain the treatment effect.



The multiple baseline design can easily show generalization across subjects, treatment conditions, or settings. Fox, et al. (1984) chose to do a multiple baseline across nonhandicapped peer confederates, and found spontaneous increases in social initiations and interactions in the treatment setting to the fourth of four nonhandicapped peers, using a single adolescent autistic girl. Additionally, after introduction of the second peer confederate, the subject showed increased social responding in a generalization setting. Odom, et al. (1985), using a multiple baseline across setting, replicated across three nonhandicapped peers confederates, failed to find generalization across settings; their design did not allow the measurement of generalization across peer confederates.

The authors of both papers were aware of the desirability of obtaining generalization, and both chose designs which made possible the measurement of generalization—in the one instance (Fox, et al.) , across peers and setting, and in the other instance, across settings.

Although there were a number of differences in design, subject, and variables, the major difference in the two studies is that Fox, et al. (1984) deliberately set about to try to obtain generalized social responding. Their design allowed them to see whether successive introduction of intervention phase generalized across peers, and whether social interactions increased in a generalization setting. Odom, et al. (1985) used a design which would have shown treatment across settings, had such occurred, but not across peer confederates.

The major weaknesses of Fox, et al. (1984) were that only one subject was used, and that there were no follow-up data. Although Odom, et al. (1985) did use more than one subject, they also failed to provide follow-up.

While Odom, et al. (1985) showed that they could increase the social interactions of handicapped preschoolers, they also showed that it was necessary to maintain the artificial nature of their treatment condition in order to maintain the interactions. While Fox, et al. (1984) did not show that their treatment effect would maintain across time, they did successfully demonstrate that increases in social responding occurred to “new” peers and generalized to a new setting.

In summary, both studies failed to address issues of maintenance of treatment effects. Additionally, Odom, et al. (1985), although they discussed generalization (like the weather), their treatment design resulted in them not “doing anything about it.” By designing their study in a way which allowed measurement of generalization across settings and peer confederates, Fox, et al. (1984) effectively added to the knowledge base of generalization.




Fox, J.J., & McEvoy, M.A. (In preparation). Assessing and enhancing generalization and social validity of social skills interventions with children and adolescents. Center for Childhood Learning and Development, East Tennessee State University, Johnson City, TN.

Fox, J.J., Gunter, P., Brady, M.P., Bambara, L., Spiegel-McGill, P., & Shores, R.E. (1984). Monograph in behavioral disorders: Severe behavior disorders of children and youth, 7, 17-26.

Hersen, M., & Barlow, D. (1976). Single case experimental designs: Strategies for studying behavior change. New York: Pergamon Press. 

Odom, S.L., Hoyson, M., Jamieson, B., & Strain, P.J. (1985). Increasing handicapped preschoolers’ peer social interactions: Cross-setting and component analysis. Journal of Applied Behavior Analysis, 18, 3-16.

Skinner, B.F. (1938). The behavior of organisms. New York: Appleton-Century.

Stokes, T.B., & Baer, D.M. (1977). An implicit technology of generalization. Journal of Applied Behavior Analysis, 10, 349-367.

Strain, P., & Timm, M. (1974). An experimental analysis of social interaction between a behaviorally disordered preschool child and her classroom peers. Journal of Applied Behavior Analysis, 7, 583-590.