1992, 58,

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR

19-36

NUMBER 1

(JULY)

PATTERNS OF RESPONDING WITHIN SESSIONS FRANCES K. MCSWEENEY AND JOHN M. HINSON WASHINGTON STATE UNIVERSITY

Rates of responding changed systematically across sessions for rats pressing levers and keys and for pigeons pressing treadles and pecking keys. A bitonic function in which response rates increased and then decreased across sessions was the most common finding, although an increase in responding also occurred alone. The change in response rate was usually large. The function relating responding to time in session had the following general characteristics: It appeared early in training, and further experience moved and reduced its peak; it was flatter for longer sessions; and it was flatter, more symmetrical, and peaked later for lower than for higher rates of reinforcement. Factors related to reinforcement exerted more control over the location of the peak rate of responding and the steepness of the decline in response rates than did factors related to responding. These within-session changes in response rates have fundamental theoretical and methodological implications. Key words: response rate, time in session, lever press, key press, treadle press, key peck, rats, pigeons

rates declined across the session, then shorter sessions might yield higher average rates of responding than longer sessions do. The effect of the independent variable on the average rate of responding would not be interpretable if sessions of different lengths were conducted for different values of the independent variable. Third, understanding the distribution of responding within sessions is essential when using within-session procedures. Within-session procedures present different aspects of an independent variable in different parts of a single session rather than in different sessions (e.g., Ettinger & Staddon, 1983; Heyman, 1983; McSweeney & Melville, 1988). These procedures are designed to eliminate the confounding effects of the large shifts in baseline response rates that can occur across sessions (e.g., McSweeney, Dougan, Higa, & Farmer, 1986; Spealman & Gollub, 1974). These procedures also save time. All values of the independent variable appear during the same session, rather than successively. Saving time is not only convenient but may also be necessary when examining the behavior of a shortlived species such as the rat. For the withinsession procedure to be used successfully, responding must be relatively constant across the session. Otherwise, changes in responding might be confounded with the independent variable, thus complicating the interpretation of the data. Finally, examining within-session patterns of responding can provide information about traditional questions that have been relatively neglected by operant psychologists. Questions

Absolute rate of responding is one of the primary dependent variables in operant psychology. Yet little is known about the way in which response rates change across experimental sessions. Investigators usually assume no important variation occurs, but this assumption is rarely tested directly. A better understanding of within-session patterns of responding may be important for several reasons. First, absolute response rates are often highly variable. A description of the distribution of responding within sessions might help to eliminate some of this variability. For example, if subjects responded erratically for the first 10 min of the session, then ignoring those responses would clarify the data. Second, experiments often confound session length with other independent variables. For example, when studying the effect of rate of reinforcement on rate of responding, the experimenter must either confound session length or number of reinforcers delivered per session with rate of reinforcement. Session length is often confounded (e.g., Catania & Reynolds, 1968) because confounding the number of reinforcers might produce systematic changes in satiation. However, allowing session length to covary with the independent variable may also yield uninterpretable average rates of responding if response rates change systematically across the session. If, for example, response This research was partially supported by NIMH Grant MH 42466. Reprints may be obtained from Frances K. McSweeney, Department of Psychology, Washington State University, Pullman, Washington 99164-4820.

19

FRANCES K. McSWEENEY and JOHN M. HINSON

20

LEEVE:R PREESS

Q16 e 012

- a

009 QS

.-0240 _6

d

-

-M. D130

--D10 son

-M.

000

#

D6

DDO

2

0

4

REa

Q12

8

a

10

12

PREESS

w

009

0

,

D240

-

OD 00

"I

-

4

W.

--

-

D13O D10 D6

I.-

0

2

4

8

10

DO

12

0O#NEN Fig. 1. Proportion of total-session responses that were emitted during successive components of the experimental sessions in McSweeney et al. (1990). The top graph presents the results for lever pressing for Noyes pellets. The bottom graph presents the results for key pressing for sweetened condensed milk. Each individual function presents

WITHIN-SESSION RESPONDING on such topics as fatigue, satiation, and warmup effects are most effectively addressed by examining within-session responding. The present paper uses data from our laboratories to examine the distribution of responding within sessions. The paper is designed to emphasize the importance of questions on this topic and to provide directions for future research, rather than to provide definitive answers. In some cases, studies were not designed to answer questions about within-session patterns of responding. Therefore, their results are suggestive rather than definitive. Lever and Key Pressing by Rats McSweeney, Hatfield, and Allen (1990) attempted to replicate a study by Bacotti (1976) that examined whether changes in the time of delivery of a postsession feeding would alter responding within the session. In this study, rats pressed levers for Noyes pellets or keys for sweetened condensed milk reinforcers delivered by multiple variable-interval (VI) 1 -min VI 1-min schedules. The components of the multiple schedule alternated every 5 min, and 12 components were delivered per session. The time at which postsession feedings were delivered was varied in five steps from 0 to 240 min after the session. Forty sessions were conducted for the first delay to postsession feedings, and 30 sessions were conducted for each subsequent delay. Neither the absolute rates of responding nor the withinsession patterns of responding varied with the delay to postsession feedings, but a consistent and pronounced pattern of within-session responding emerged for both responses. That is, absolute response rates increased for the first 20 min of the session and then decreased. Figure 1 replots the within-session patterns of responding reported by McSweeney et al. (1990). It presents the proportion of the totalsession responses emitted during successive components within the session. Proportions were calculated by dividing the number of responses emitted during a component by the total number of responses emitted during the session. The plotted proportions are the means of the proportions calculated over all subjects

21

and over the last five sessions for which each delay was conducted. The top graph presents the results for lever pressing for Noyes pellets. The bottom graph presents the results for key pressing for sweetened condensed milk. The different functions present the results for different delays to postsession feedings. Proportions have been presented here and throughout this paper so that the pattern of responding within the session will be independent of differences in absolute response rate. (Absolute values may be obtained by consulting the cited references from which the data were derived or by writing to either author.) The presentation of proportions does not distort the data by obscuring any other functional relations. The results in Figure 1 also represent the results for individual subjects. Figure 2 presents the proportion of total-session responses emitted by individual subjects during successive components for lever (top) and key (bottom) pressing. The proportions were calculated over the last five sessions for which the 60-min postsession feeding delays were conducted. Data for the 60-min postsession feedings were arbitrarily selected for presentation. Figure 3 shows that the results presented in Figure 1 also represent the results for individual sessions. It presents the proportion of totalsession responses emitted by the first subject during successive components for each of the last five sessions during which the 60-min delay to postsession feedings was presented. Results are presented for lever (top) and key (bottom) pressing. Data for the 1st subject were arbitrarily selected for presentation. The functions reported in Figures 1, 2, and 3 are orderly. In all cases, response rates increased for approximately the first 20 min (four 5-min components) of the session and then decreased. This function was reported for both responses, both reinforcers, all subjects, and all sessions. The changes in response rates across the session were large. For example, the rate of key pressing averaged 15.7 and 27.5 responses per minute for the first and 12th components, respectively. It averaged 70.0 responses per minute at its peak in the fourth component. Lever pressing averaged 6.8 and

the results for subjects responding at a particular delay to postsession feedings. The results are for the mean of all subjects averaged over the last five sessions for which each delay to postsession feeding was presented. The delay values (following D) are in minutes.

FRANCES K. McSWEENEY and JOHN M. HINSON

22

LEVEFs PRFS S 0.15 r-

Ei t

0.10

Ar-

06 0.05

L.......- -SS - ---8S4 -

-

S3 S2

S1

0.00 0

4

2

6

8

10

12

COMPONENT

KERY PRFES ES 0.25

E 0.20

0.15

Y 0.10

E 0.06

84 83

82 Si

- -

-L

0.00 0

2

4

6

1

1

8

10

, 12

COMPONENT Fig. 2. The proportion of total-session responses emitted by individual subjects during successive components in McSweeney et al. (1990). The top graph presents results for lever pressing; the bottom graph presents results for key pressing. Each function presents the results for a single subject. The results are the means of the last five sessions for which the 60-min delay to postsession feeding was conducted.

WITHIN-SESSION RESPONDING

23

LEEVER PRE8SS 020 r

aQ16

Q10

E

06

0ao L 0

2

4

8 8 OOMPONENT

10

12

KEEY PRES S 020

I

Q16

Q10 F

a

06 F - a

aoo

A

0

2

4

8 6 OOMPONENT

10

12

Fig. 3. The proportion of total-session responses emitted during successive components for individual sessions in McSweeney et al. (1990). The top graph presents results for lever pressing; the bottom presents results for key pressing. Each individual function presents the results for one of the last five sessions for which the first subject (SI in Figure 2) responded during the 60-min delay to postsession feedings. (Reprinted with permission.)

FRANCES K. McSWEENEY and JOHN M. HINSON

24

24

0.3

02

IS

\

cJ

Q

0.1

0.0

L

I

0

1

I

I

4 3 2 BGHT-MINUTE INTERVALS

6

I

I

.-. 8240 8120 - --860 -__-- 810

Fig. 4. The proportion of total-session treadle presses emitted during successive 8-min intervals during the session when pigeons responded on positive conditioned-suppression procedures. Each function presents the results for a particular stimulus duration (in seconds). Results have been averaged over the last five sessions for which each duration was presented and are those for the mean of all subjects.

session treadle presses that were emitted during successive 8-min intervals during the session. The results are those for the mean of all subjects responding during the last five sessions for which each stimulus duration was presented. Each individual function presents the results for a particular stimulus duration. Results have been presented for 8-min intervals, Treadle Pressing by Pigeons instead of for the 5-min intervals plotted in Figure 4 presents the results obtained when the earlier figures, because the results were pigeons pressed treadles for mixed grain re- more orderly when plotted for longer periods. Figure 5 presents the results obtained for inforcers delivered by a positive conditionedsuppression procedure. Pressing a treadle pro- individual subjects responding when the stimduced reinforcers according to a VI 1-min ulus was 10, 60, 120, or 240 s long. Each graph schedule. From time to time, a red light ap- presents the results for a particular stimulus peared on the response key located directly duration, and each function presents the reabove the treadle. The stimulus was followed sults for an individual subject. Again, results by a response-independent reinforcer. Stimuli have been averaged over the last five sessions for which each stimulus duration was prewere presented according to a VI 1-min schedule, and no reinforcers were delivered during sented. The results for the 120-s stimulus were the stimulus. The duration of the stimulus var- variable from subject to subject. Otherwise, ied across conditions in five steps from 10 to response rates increased and then decreased 240 s. Each stimulus duration was presented during the session. They reached a peak apfor 40 sessions. All sessions ended when 40 proximately 20 min after the beginning of the min had elapsed. Reinforcers were 5 s of access session. Again, the change in response rates across the session was large. Subjects emitted to mixed grain. Figure 4 presents the proportion of the total- an average of 16% or 17% of their total re-

11.9 responses per minute during the first and 12th components, respectively. It averaged 19.4 responses per minute at its peak in the fourth component. This function appeared even though the distribution of reinforcers programmed for delivery by the multiple VI VI schedule was flat across the session.

WITHIN-SESSION RESPONDING

25

630 a

10 S 04 r

064 as

e ai

b 02f-

I0.8

1

2

a

4

a

00

1

eOW-4M4E HTERAL8

4

8

6

4

_

e-8l

-81

EI r W E NrRAL

120 S 04

2

-

240 a 04r a-

box as

00

I

1: t

8 Eu4r-*4E INL/M8 2

a-

--844 88

---

0

1

2

ES NI

8 IN I

4

6

---621

VVALM

Fig. 5. The proportion of total-session treadle presses emitted during successive 8-min intervals during the session. Each graph presents the results for one particular stimulus duration. The functions represent the results for individual subjects. The results have been averaged over the last five sessions for which a particular stimulus duration was presented. sponses in the first and last fifths, but they emitted 24% of their total responses in the third fifth.

Key Pecking by Pigeons Figure 6 presents the proportion of totalsession key pecks emitted during successive 5-min intervals by pigeons pecking keys for food reinforcers delivered on concurrent VI VI schedules. The left side of the figure presents the results obtained for subjects responding on two-key concurrent schedules. Each schedule was presented for 14 daily 1-hr sessions (see Hinson & Staddon, 1983, for further details). The right side of the figure presents the results obtained for subjects responding on Findley

(1958) concurrent schedules. In this experiment, a yellow changeover key controlled a schedule key signaled by either red or green. Each schedule was presented for 20 30-min sessions (see Hinson & Higa, 1989, for a description of the apparatus). In both experiments, the reinforcer was 3 s of access to mixed grain. Each graph presents the results for the mean of the 3 subjects responding calculated over the last seven sessions for which the schedule was presented. The two functions plotted on each graph present the results for the two components of each schedtle. Figure 7 presents results for individual subjects. Again, the proportion of the total-session responses emitted during successive 5-min in-

FRANCES K. McSWEENEY and JOHN M. HINSON

26

FlNIDLEEY VI 30 V1 30

TW/VC0-KE-Y VI 60 VI130 £L12 a10

aIo

0 //a16

/a

10051

1a am aOD

2

c0

4

a

a

10

Goo c

12

8 2 #N nWE

1

FE-*f Nl

i

8 4 O n8

6

VI 30 VI 90

VI 240 VI C0 amo

Q12

aio

IceD

ata /

I ,l_

6

Om

aio

I

I

~

.

s

1

4

8

000

ow 2

0042

4

0

a

10

1

12

2

a

a

CAiI

[c

010

VI 90V 1 30

VI 90 VI 1 80 020

-

/

aoos

'I

~

~

f

06 Q_ ax

012

OD

000 0

2

4

a

a

n*MEJF 3NrMA

t0

12

1

S 4 2 FIV MNMHi

8

6

WITHIN-SESSION RESPONDING tervals for subjects responding on concurrent schedules is shown. Each graph presents the results for a single subject responding on the two-key concurrent VI 90-s VI 180-s or the Findley concurrent VI 30-s VI 90-s schedule. The two functions plotted on each graph represent the results for the two components of each schedule. Again, data have been averaged over the last seven sessions for which the schedule was available. The bitonic function appears for all three two-key concurrent schedules and for each individual subject, although it is fairly flat for Subject 1. The ascending limb of the function appears for all schedules and subjects for the Findley concurrent schedules. It is not apparent in Figure 7 for the first component for Subject A because the function reached its peak during the first 3 min. Therefore, Figure 7 does not present a fine enough resolution of the data for the ascending limb to be apparent. The descending limb did not appear clearly for all Findley concurrent schedules or for individual subjects. The differences between the results for twokey and Findley concurrent schedules should not be interpreted too strongly. The descending limb may have failed to appear for the Findley concurrent schedules for minor procedural reasons (e.g., because sessions were too short or rates of reinforcement were too high), as well as because of more fundamental differences between the behavior governed by Findley and two-key concurrent schedules. Factors That Control Within-Session

Responding Session length. The factors that control the changes in response rates across the session need to be determined. In this regard, the experiment for rats pressing keys was repeated varying session length (McSweeney, in press, Experiment 2). All procedural details were similar to those listed earlier for key pressing except that the following session lengths were conducted in the following order (the component duration used to produce each session

27

length appears in parentheses after the session length): 60 min (300 s), 30 min (150 s), 90 min (450 s). Session length was changed by varying the length of the components while presenting a constant 12 components per session. Subjects responded on a multiple VI 1 -min VI 1 -min schedule for 30 sessions for each session length. Figure 8 presents the proportion of totalsession responses emitted during successive components for each session length. The individual functions present the results for different session lengths. Responses have been averaged over all subjects and over the last five sessions for which each session length was conducted. (Results for individual subjects resemble those in Figure 8 and may be found in McSweeney, in press.) The peak response rates occurred during later components for shorter than for longer sessions. Because components were shorter for shorter sessions, peak response rates occurred approximately the same number of minutes after the beginning of the session for all session lengths. Responding peaked between 12.5 and 22.5 min (Components 6 to 9) for the 30-min sessions, 15 and 20 min (Component 4) for the 60-min sessions, and 15 and 37.5 min (Components 3 to 5) for the 90-min sessions. The functions reported in Figure 8 were flatter for longer sessions. The proportion of responses emitted during the components varied from .02 to .11 for the 30-min sessions. They varied from .05 to .11 for the 60- and 90-min sessions. This may be a real effect of session length, or it may be a statistical artifact. Suppose, for example, that subjects responded at a very slow rate during the first minute of the session regardless of session length. This slow responding would influence the reported proportion of first-component responses more for shorter than for longer sessions. Because components are shorter for shorter sessions, responding during the first minute would exert a larger influence on responding during the first component for the shorter sessions.

Fig. 6. Proportion of total-session key pecks emitted during successive 5-min intervals for each component of twokey (left) or Findley (right) concurrent VI VI schedules. The results for the two-key concurrent schedules were previously reported by Hinson and Staddon (1983). Each graph presents the results for the mean of the 3 subjects responding in that experiment. The two functions present the results for the two components. The solid line presents the results of the component listed first; the dashed line, the component listed second. The concurrent schedules are reported in seconds. Proportions have been calculated over the last seven sessions for which each schedule was available.

FRANCES K. McSWEENEY and JOHN M. HINSON

28

F=INI DL-E=Y

SUBJE-CT

8UBJEGCT A

1 020

16

r

0.12

/0

-4..

2

4

le

016

eA-&

aI aow

2

s 10 a NAE NnGTB

4

a UEBJ E=-:.-C

ao0

12

0

1

m20

6

a

r

Ra-

0.16 0,12

.

6

2

ato

t

6

rE

--f

N

r

.

a106

amf 1:

aoo

0

2

a a nVE-MNUE NIVX 4

SUBJEC)T

10

o

12

co

1

4 2 FWE-MWE NHT1WAU

SU6_JE'DT

3

016

C

020

_

~

~~ _- _

0.16

b O.

4(/

0.10

I

CM0o

-

- --P2

0 0 L 0

I

2

.

I IA. 10

I

a a nE4NE NK'VVL 4

12

^

0

P1

1

2

4

RYWAEHiWllI

6

6

WITHIN-SESSION RESPONDING The functions were also more symmetrical around the middle of the session for the intermediate session length (60 min) than for longer and shorter sessions. Wilcoxon signed rank tests showed that the proportion of responses emitted during the first and second halves of the sessions differed significantly (p < .05) for the 30-min (p = .04) and 90-min (p = .04) sessions, but not for the 60-min sessions (p = .08). More responses were emitted during the second than during the first half of the session when sessions were 30 min long. More responses were emitted during the first than during the second half when sessions were 90 min long. A one-way within-subject analysis of variance confirmed that the absolute rates of responding emitted during the entire session also changed with session length, F(2, 8) = 4.87, p < .05. The mean response rate emitted during the 60-min sessions (64.8 responses per minute) was significantly (p < .05) greater than that emitted during the 30-min sessions, t(4) = -2.84, M = 36.9 responses per minute, or 90-min sessions, t(4) = 4.11, M = 47.5 responses per minute. Rate of reinforcement. To determine the effect of the programmed rate of reinforcement on the changes in response rates within sessions, the experiment for rats pressing levers was repeated varying the rate of reinforcement (McSweeney, in press, Experiment 1). Subjects responded for Noyes pellets on multiple schedules in which the components alternated every 5 min. The following schedules were presented in the following order: multiple VI 30 s VI 30 s, multiple VI 2 min VI 2 min, multiple VI 4 min VI 4 min, multiple VI 15 s VI 15 s, and multiple VI 1 min VI 1 min. Subjects responded on each of these schedules for 30 sessions. Each session ended when 12 components had been presented. Figure 9 presents the proportion of totalsession responses emitted during successive components of the multiple schedules. The individual functions present the results for the

Q16

r

012

-

29

ODO

I--- MOO QOO

-

0

2

a 10 8 8UOEB8IVE OOMPONB4T8 4

MOo

12

Fig. 8. The proportion of total-session responses emitted during successive components for the mean of all subjects responding in Experiment 2 of McSweeney (in press). Each function presents the results for a different session length. Results have been averaged over the last five sessions for which each session length was available. Session lengths (values after M) are in minutes. (Reprinted with

permission.)

different multiple schedules. Results have been averaged over all subjects and over the last five sessions for which each schedule was conducted. (Results for individual subjects are similar to those for the mean of all subjects and appear in McSweeney, in press.) The functions became flatter as the programmed rate of reinforcement decreased. For example, proportions varied from .02 to .15 for the multiple VI 15-s VI 15-s schedule, but from only .07 to .11 for the multiple VI 4-min VI 4-min schedule. The functions also became more symmetrical around the middle of the session as the programmed rate of reinforcement decreased. Subjects emitted significantly (p < .05) more responses during the first half of the session than during the second half for the VI 15-s (p = .04), VI 30-s (p = .04), and VI 60-s (p = .04) schedules, as assessed by Wilcoxon signed rank tests. The proportion of responses emitted during the two halves did not differ signifi-

Fig. 7. Proportion of total-session key pecks emitted by individual subjects during successive 5-min intervals for the two-key concurrent VI 90-s VI 180-s (left) and the Findley concurrent VI 30-s VI 90-s (right) schedules. Each graph presents the results for an individual subject. The two functions present the results for the two components. The solid line presents the results for the shorter VI schedule; the dashed line, for the longer VI schedule. Proportions have been averaged over the last seven sessions for which each schedule was available.

FRANCES K. McSWEENEY and JOHN M. HINSON

020

| Q16

5 0Q10

.V*

106

- - -

-

oflo

"

40

V1120

"WV110

I

0

2

4

8

10

12

SUCCE81VE COATSNENT8 Fig. 9. Proportion of total-session responses emitted during successive components for the mean of all subjects responding in Experiment 1 of McSweeney (in press). Each function presents the results for a particular VI schedule. Results have been averaged over the last five sessions for which each schedule was presented. Schedule values are in seconds. (Reprinted with permission.)

cantly for the VI 2-min (p = .07) or the VI 4-min (p = .89) schedules. The peak rate of responding occurred later in the session as the programmed rate of reinforcement decreased (Figure 9). Subjects emitted the highest proportion of their total responses during the first or second component of the multiple VI 15-s VI 15-s schedule. Subjects emitted the highest proportion of their total responses during the sixth component of the multiple VI 4-min VI 4-min schedule. In this experiment, the average rate of responding (responses per minute) emitted over the entire session increased with increases in the programmed rate of reinforcement up to 120 reinforcers per hour (multiple VI 30-s VI 30-s schedule). Response rates decreased with a further increase in reinforcement rate to 240

reinforcers per hour. These results appear in the top left graph in Figure 10 (taken from McSweeney, in press) and are labeled "mean." The results are those for the mean of all subjects responding over the last five sessions for which each schedule was available. Such decreases in response rates at high rates of reinforcement are frequently observed (e.g., Dougan & McSweeney, 1985; McSweeney & Melville, 1991). The question arises whether response rates would increase as a monotonic function of rate of reinforcement if factors that changed over the session (e.g., satiation, fatigue) were prevented. The graphs in Figure 10 answer this question. They present the mean rates of responding emitted during the first, third, ninth, and 12th components plotted as a function of

--4

Fig. 10. Rates of responding in Experiment 1 of McSweeney (in press) plotted as a function of programmed rates of reinforcement. The top left graph presents rates of responding averaged across the entire session. The other graphs present response rates for the first, third, ninth, and 12th components. All results are for the mean of all subjects averaged over the last five sessions for which each schedule was available. (Reprinted with permission.)

WITHIN-SESSION RESPONDING

NIST

NA1 E 600

8o

F

I~~ t

601

I

31

i~~~~

60

I~~~1

i/

/

80 f 401i

* 40

/

-I !

w

201

20

10 60

160 100 200 R_BIn AT RATE

10, 0

260

1

--

60

200

260

RATE

1IN11tXTH

60

70 r

80

I

I /~~~~/\ 40

/

w a60

s0o

_

so 10

20

N

20a

L

IO

\0

I

10

60

100

160

4FqCRM

200 RATE

T V/ELFT7 H 86[

i-o

II]

IJ3

-

W Ul

/

40 r

§26

160

rW

THiPRD

80

L_ _-_A_-

.

100

'

T\\ /so~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~:

0

60

100

160

fGVMWT RATE

200

260

FRANCES K. McSWEENEY and JOHN M. HINSON

32

0.12 i

VI 30-8

ala r

a1o t

I outF

0.12 1

0o0 F

*1' ~/\eX' k

.

\

004 1I &72

\

-- -

LAST

000

8

10

18

OCMNBNT 60 MINUT Es .18

0

-F~ FR

I

ra12 ld.

%

2

4

a cka

8

10

8E8826

12

Fig. 12. The mean proportion of total-session responses emitted during successive components of a multiple VI VI schedule. The solid line presents the mean proportions calculated over Sessions 25 to 30 for which that schedule was presented. The dashed line presents the mean proportions calculated over Sessions 55 to 60. The results are those for the mean of all 5 subjects responding on this schedule.

ted during successive components of experimental sessions. In both graphs, the first ses00X sion had been preceded only by earlier sessions I of shaping. The data are those for the mean 0a0SF of all subjects. A bitonic function was present even during the first recorded session (Figure - - - Lm# -'UT' 11). The effect of further training was to reoo 16 10 a o duce the difference between the highest and OMCNE lowest rates of responding and to move the Fig. 11. The proportion of total-sessi(on responses peak of the function to the point that was apemitted during successive components in E xperiment 1 propriate for that schedule and session dura2 (bottom axes) off McSweeney and 4I

/I

-

Experiment (top axes) (in press). Results are presented for the Eirst recorded session of training (circles) and for the meain of the last five sessions of training (triangles) for the mean of all subjects responding on the multiple VI 30-s VtI 30-s schedule (top graph) or the 60-min session length (l ottom graph). (Reprinted with permission.)

the programmed rate of reinforcer,nent. The results are those for the mean of a.11 subjects responding during the last five se ssions for which each schedule was availab le. These graphs show that response rate increased monotonically with increases in the rate of reinforcement during the first comp onent, but response rates declined at the highLest rate of reinforcement for all other compon ents. Experience. The present bitonic functions are apparent early in training and persist in spite of much experience with the flot dFitr bution of reinforcers across the sessi 11, taken from McSweeney (in press;), presents the proportion of total-session respe nses emit-

t

ion.

Figure 12 shows that the function changed little even when 60 sessions were conducted. It compares the mean proportion of responses emitted during successive components averaged over Sessions 25 to 30 and over Sessions 55 to 60 when rats pressed levers for Noyes pellets delivered by a multiple VI 1-min VI 1 -min schedule. Components were 150 s long, and 12 components were presented per session. The results are those for the mean of 5 subjects. (The results for the mean closely represent those for individual subjects.) The function changed little between Sessions 25 to 30 and 55 to 60. Therefore, the function persisted in spite of extensive exposure to a flat distribution of reinforcement across the session. GENERAL DISCUSSION The present results show that responding changed systematically within sessions when

WITHIN-SESSION RESPONDING subjects responded on several experimental procedures. Response rate frequently increased to a peak and then decreased. Increases in response rates were also found alone (see Figures 6 and 7). These changes in response rates occurred for several responses (key press, lever press, key peck, treadle press), procedures (multiple schedules, concurrent schedules, positive conditioned suppression), session lengths (30 to 90 min), and rates of reinforcement (15 to 240 reinforcers per hour). When response rates changed, the changes were also found for all subjects and sessions. The changes appeared in steady-state behavior as well as for responding early in training. The changes in response rates across the session appeared in spite of, rather than because of, the distribution of reinforcers across the session. The present VI schedules program a flat distribution of reinforcers across the session. Therefore, any changes in this distribution must be produced by changes in the rates of responding, not vice versa. The distribution of reinforcers may have influenced the bitonic function, however. The difference between the highest and lowest response rates emitted during the session decreased from the first to the 30th session of training (Figure 11). The flat distribution of reinforcers may have contributed to this decrease. However, the unchanging distribution of reinforcement within sessions never fully controlled the distribution of responding. Response rates changed systematically within sessions even after 60 sessions of exposure to a flat distribution of reinforcement (Figure 12). To the best of our knowledge, this function has not been previously reported, but several other studies may be related. Two variables have traditionally described performance decrements in similar situations: fatigue (e.g., Muscio, 1921) and satiation (e.g., Reese & Hogenson, 1962). Two variables have also described results similar to the ascending limb of the function. First, priming has been observed when electrical brain stimulation is used as the reinforcer. Priming refers to the fact that subjects may not respond until some free reinforcers have been given (e.g., Olds, 1956; Olds & Milner, 1954). Second, warm-up has been reported, with different definitions, in many studies, including those on the following conditioning phenomena: discriminated avoidance (e.g., Foree & LoLordo, 1970; Hoffman, Fleshler, & Chorny, 1961), Sidman avoidance

33

(e.g., Powell, 1970; Wertheim, 1965), punishment (e.g., Azrin, 1960; Azrin, Holz, & Hake, 1963; Hake & Azrin, 1965; Hake, Azrin, & Oxford, 1967), shock-elicited aggression (Hutchinson, Renfrew, & Young, 1971; Lyons & Ozolins, 1970), and conditioned emotional response procedures (e.g., Rubin & Brown, 1969). The early trials for such tasks as psychophysical procedures (e.g., Hodos, Leibowitz, & Bonbright, 1976) or generalization tests (e.g., Griffin & Stewart, 1977) may also be discarded as warm-up trials. The relation between the present results and these historical observations is not known. Fatigue has not been a useful concept when describing the literature on human performance decrements (e.g., Muscio, 1921). It includes the effects of too many variables (e.g., effects at the neural, muscular, and central levels; effects resulting from overload with those resulting from underload). Except for warm-up for avoidance (e.g., Hineline, 1978a, 1978b), neither priming nor warm-up has been studied in systematic detail. However, what is known suggests that neither procedure is related to the present functions. For example, priming is not found for all subjects. The ascending limb of the present function is found reliably for all subjects under any conditions that produce it. The duration of warm-up for avoidance is highly variable across subjects (e.g., Badia, Culbertson, & Lewis, 1971). The ascending limb of the present function ends at approximately the same time for all subjects. Warm-up for avoidance may also fail to appear in measures of response rate. Warm-up for avoidance refers to the fact that subjects may receive a large proportion of their total session shocks in the first minutes of a session. Interestingly, large changes in the number of shocks received may be reported in the absence of any change in response rate (e.g., Leander, 1973; Powell & Peck, 1969). The present effect is just the opposite. It represents a change in the rate of responding in the absence of a change in the rate of reinforcement. Regardless of its relation to earlier studies, the present function has important theoretical and methodological implications. To begin with, it suggests that caution should be used when confounding session length with an independent variable. As argued earlier, many studies require that either session length or another variable (e.g., number of reinforcers

FRANCES K. McSWEENEY and JOHN M. HINSON delivered per session) must be confounded with the independent variable (e.g., rate of reinforcement). If session length cannot be safely confounded, then operant studies will be more difficult to conduct in the future. The present results show that session length cannot be safely confounded, at least when rats press keys. Rates of responding changed significantly with moderate changes in session length. For example, the overall response rate almost doubled when the session was lengthened from 30 min (M = 36.9 responses per minute) to 60 min (M = 64.8 responses per minute). Such large changes in response rate would seriously confound the interpretation of the results. Second, the present results imply that withinsession procedures should be used with caution. Response rates change systematically across experimental sessions. Therefore, experiments must be designed to avoid confounding the effects of these changes with the effect of the independent variable. Limitations on the use of within-sessions procedures cannot be precisely specified at this time. The generality of the present results is not known. Therefore, within-session procedures might be safely used in some experiments but not in others. The limitations will undoubtedly also depend on many aspects of the experimental design. For example, the inclusion of timeouts during the session may change the distribution of responding within sessions in unknown ways. Third, the present results imply that the distribution of responding within sessions has complicated the answer to some theoretical questions. Figure 10 shows that rates of responding emitted at the beginning of the session increase monotonically with increases in the rate of reinforcement, confirming some theories (e.g., Herrnstein, 1970). Responding later in the session increases up to a point and then decreases, confirming other theories (e.g., Baum, 1981; Staddon, 1979). It remains to be seen whether within-session patterns of responding have confounded the effects of other independent variables, such as reinforcer size or delay. The size and reliability of the present function, as well as its resistance to the flat distribution of reinforcement, suggest that the function is controlled by theoretically important processes. These processes should be identified by studying the factors that control the form of the function. The present results provide

limited information on this topic that should be verified by further experiments. To begin with, the results suggest that the bitonic function reported in several of the figures (e.g., Figures 1 and 3) is not a continuous function but is composed of two independent limbs. For example, Figures 6 and 7 show that the ascending limb can appear without the descending limb. Figures 6 and 7, suggest that the descending limb of the function may fail to appear when sessions are short. Sessions were only 30 min long for the Findley concurrent schedules in Figures 5 and 6, and the descending limb of the function sometimes failed to occur during these procedures. The present results suggest that factors related to reinforcement (e.g., satiation, priming) exert more control over the peak of the function and the steepness of the descending limb than factors related to responding (e.g., fatigue, warm-up). Most of the present data cannot separate these variables because higher rates of reinforcement usually covaried with higher rates of responding. However, Figure 8 presents data critical to separating the effects of these variables on the peak of the function. Key pressing for sweetened condensed milk peaked after approximately 20 min (20 reinforcers) regardless of session duration. This was true even though the number of emitted responses was very different for different session lengths. Subjects emitted a mean of 851.5 responses in the first eight components of the session (20 min) for the 30-min sessions, a mean of 1,292.9 responses in the first four components (20 min) for the 60-min sessions, and a mean of 1,061.0 responses in the first three components (22.5 min) for the 90-min sessions. This means that responding reached a peak after 20 reinforcers or 20 min, regardless of the number of responses emitted. Data critical to separating the effects of reinforcement from the effects of time are found in Figure 9. It shows that the ascending limb of the function is steeper for higher rates of reinforcement, suggesting that reinforcement exerts stronger control over this limb than the passage of time alone. Figure 9 presents data critical to separating the effects of responding and reinforcement on the descending limb of the function. One comparison is critical. Subjects responded faster but collected fewer reinforcers on the multiple VI 30-s VI 30-s schedule (M = 32.6 responses

WITHIN-SESSION RESPONDING per minute) than on the multiple VI 15-s VI 15-s schedule (M = 18.5 responses per minute). Figure 9 shows that the descending limb of the function was steeper for the multiple VI 15-s VI 15-s schedule than for the multiple VI 30-s VI 30-s schedule. Therefore, reinforcement may be more important than responding in controlling the descending limb. The results presented in this paper raise more questions than they answer. The factors that control the changes in response rates across sessions should be more carefully determined. For example, the sensitivity of the changes to the distribution of reinforcers across the session should be studied. The contributions of response-related and reinforcement-related variables should be more carefully separated. The generality of the changes in response rates for different schedules, species, reinforcers, responses, and procedures should be determined. The present data suggest that changes may be quite general because they report these changes for several different procedures, responses, reinforcers, and species. However, responding may not always change systematically across the session (e.g., Bloomfield, 1967). The factors that produce the different results should be determined. The implications of the present functions for theories of operant behavior should also be investigated. Most theories of operant behavior use rate of responding as their dependent variable. Because the present results show that the average rate of responding across the session masks strong regularities in behavior at a more molecular level, the present results suggest that many theories of operant behavior may need revision. To give just one example, it is generally accepted that changing the rate of reinforcement changes the average rate of responding over the session. But Figure 9 shows that this is an incomplete description. Changing the rate of reinforcement also changes the distribution of responding within the session, even when differences in the absolute rates of responding have been eliminated by reporting proportion of total responses rather than response rates. Newer theories must be formulated to capture these regularities.

REFERENCES Azrin, N. H. (1960). Effects of punishment intensity during variable-interval reinforcement. Journal of the Experimental Analysis of Behavior, 3, 123-142.

35

Azrin, N. H., Holz, W. C., & Hake, D. F. (1963). Fixed-ratio punishment. Journal of the Experimental Analysis of Behavior, 6, 141-148. Bacotti, A. V. (1976). Home cage feeding time controls responding under multiple schedules. Animal Learning & Behavior, 4, 41-44. Badia, P., Culbertson, S., & Lewis, P. (1971). The relative aversiveness of signalled vs. unsignalled avoidance. Journal of the Experimental Analysis of Behavior, 16, 113-121. Baum, W. M. (1981). Optimization and the matching law as accounts of instrumental behavior. Journal of the Experimental Analysis of Behavior, 36, 387-403. Bloomfield, T. M. (1967). Some temporal properties of behavioral contrast. Journal of the Experimental Analysis of Behavior, 10, 159-164. Catania, A. C., & Reynolds, G. S. (1968). A quantitative analysis of the responding maintained by interval schedules. Journal of the Experimental Analysis of Behavior, 11, 327-383. Dougan, J. D., & McSweeney, F. K. (1985). Variations in Herrnstein's ro as a function of alternative reinforcement rate. Journal of the Experimental Analysis of Behavior, 43, 215-223. Ettinger, R. H., & Staddon, J. E. R. (1983). Behavioral competition, component duration and multiple schedule contrast. Behavioural Analysis Letters, 2, 31-38. Findley, J. D. (1958). Preference and switching under concurrent scheduling. Journal of the Experimental Analysis of Behavior, 1, 123-144. Foree, D. D., & LoLordo, V. M. (1970). Signalled and unsignalled free-operant avoidance in the pigeon. Journal of the Experimental Analysis of Behavior, 13, 283290. Griffin, P., & Stewart, D. J. (1977). Line-orientation generalization following signalled-reinforcer training. Journal of the Experimental Analysis of Behavior, 25, 151-160. Hake, D. F., & Azrin, N. H. (1965). Conditioned punishment. Journal of the Experimental Analysis of Behavior, 8, 279-293. Hake, D. F., Azrin, N. H., & Oxford, R. (1967). The effects of punishment intensity on squirrel monkeys. Journal of the Experimental Analysis of Behavior, 10, 95107. Herrnstein, R. J. (1970). On the law of effect. Journal of the Experimental Analysis of Behavior, 13, 243-266. Heyman, G. M. (1983). A parametric evaluation of the hedonic and motoric effects of drugs: Pimozide and amphetamine. Journal of the Experimental Analysis of Behavior, 40, 113-122. Hineline, P. N. (1978a). Warmup in avoidance as a function of time since prior training. Journal of the Experimental Analysis of Behavior, 29, 87-103. Hineline, P. N. (1978b). Warmup in free-operant avoidance as a function of the response-shock-shock-shock interval. Journal of the Experimental Analysis of Behavior, 30, 281-291. Hinson, J. M., & Higa, J. J. (1989). Discrete and continuous measures of dimensional stimulus control. Journal of the Experimental Analysis of Behavior, 51, 199-214. Hinson, J. M., & Staddon, J. E. R. (1983). Hill-climbing by pigeons. Journal of the Experimental Analysis of Behavior, 39, 2-47. Hodos, W., Leibowitz, R. W., & Bonbright, J. C., Jr.

36

FRANCES K. McSWEENEY and JOHN M. HINSON

(1976). Near-field visual acuity of pigeons: Effects of head location and stimulus luminance. Journal of the Experimental Analysis of Behavior, 25, 129-141. Hoffman, H. S., Fleshler, M., & Chorny, H. (1961). Discriminated bar-press avoidance. Journal of the Experimental Analysis of Behavior, 4, 309-316. Hutchinson, R. R., Renfrew, J. W., & Young, G. A. (1971). Effects of long-term shock and associated stimuli on aggressive and manual responses. Journal of the Experimental Analysis of Behavior, 15, 141-166. Leander, J. D. (1973). Shock intensity and duration interactions on free-operant avoidance behavior. Journal of the Experimental Analysis of Behavior, 19, 481490. Lyons, D. O., & Ozolins, D. (1970). Pavlovian conditioning of shock-elicited aggression: A discrimination procedure. Journal of the Experimental Analysis of Behavior, 13, 325-331. McSweeney, F. K. (in press). Rate of reinforcement and session duration as determinants of within-session patterns of responding. Animal Learning & Behavior. McSweeney, F. K., Dougan, J. D., Higa, J., & Farmer, V. A. (1986). Behavioral contrast as a function of component duration and baseline rate of reinforcement. Animal Learning &r Behavior, 14, 173-183. McSweeney, F. K., Hatfield, J., & Allen, T. M. (1990). Within-session responding as a function of post-session feedings. Behavioural Processes, 22, 177-186. McSweeney, F. K., & Melville, C. L. (1988). Positive contrast as a function of component duration using a within-session procedure. Behavioural Processes, 16, 2141. McSweeney, F. K., & Melville, C. L. (1991). Behavioral contrast as a function of component duration for leverpressing using a within-session procedure. Animal Learning & Behavior, 19, 71-80. Muscio, B. (1921). Is a fatigue test possible? British Journal of Psychology, 12, 31-46.

Olds, J. (1956). Runway and maze behavior controlled by basomedial forebrain stimulation in the rat. Journal of Comparative and Physiological Psychology, 49, 507512. Olds, J., & Milner, P. (1954). Positive reinforcement produced by electrical stimulation of septal area and other regions of the rat brain. Journal of Comparative and Physiological Psychology, 47, 419-427. Powell, R. W. (1970). The effect of shock intensity upon responding under a multiple-avoidance schedule. Journal of the Experimental Analysis of Behavior, 14, 321329. Powell, R. W., & Peck, S. (1969). Running-wheel activity and avoidance in the mongolian gerbil. Journal of the Experimental Analysis of Behavior, 12, 779-787. Reese, T. W., & Hogenson, M. J. (1962). Food satiation in the pigeon. Journal of the Experimental Analysis of Behavior, 5, 239-245. Rubin, H. B., & Brown, H. J. (1969). The rabbit as a subject in behavioral research. Journal of the Experimental Analysis of Behavior, 12, 663-667. Spealman, R. D., & Gollub, L. R. (1974). Behavioral interactions in multiple variable-interval schedules. Journal of the Experimental Analysis of Behavior, 22, 471-481. Staddon, J. E. R. (1979). Operant behavior as adaptation to constraint. Journal of Experimental Psychology: General, 108, 48-67. Wertheim, G. A. (1965). Some sequential aspects of IRTs emitted during Sidman-avoidance behavior in the white rat. Journal of the Experimental Analysis of Behavior, 8, 9-15.

Received July 19, 1991 Final acceptance December 10, 1991