Identify 2 reasons why an investigator should be concerned about trends in the data that have no obvious explanation and what a prac

CHAPTER NINE

Steady States and Transitions

THE STEADY-STATE STRATEGY Collecting Repeated Measures Comparing States of Responding The Risk of Extraneous Variables Summary

STEADY STATES Definition Uses

Evaluates Measurement Decisions Reveals the Influence of Conditions Evaluates Experimental Control Facilitates Experimental Comparisons

Identification Trends Range Cycles

Criteria Uses Statistical Graphical Nondata

Establishing Stable Responding

TRANSITIONS Transition States Transitory States Identification Making Phase Change Decisions

191

I

192 9 . S1' EADY STATES AND TRANsrr

IONs

1,.

Manipulation of new variables will often produce changes, but tn order to describe tbe cbanges, we must be able to specify the baseline

0111 wbicb tbey occurred. -Murray Sidman

1HE STEADY-STATE STRATEGY

Collecting Repeated Measures

Let us suppose that we have defined a response class, selected a dimensional quantity, set up observation procedures, and are ready to sta~ collecting data under a baseline ( control) condition. The first graphed data point summarizing responding during a session will tell us something we never knew before, but it will only make it obvious that one data point does not tell us very much about what responding looks like under this condition. In particular, we would not know whether this value is typical of what we should expect under this base­ line condition.

The only way to answer this question is to observe for another session. What we are likely to find is that our second data point is not the same as the first. Our question then becomes: "Which of these two values is more representative of the impact of this phase?" Again, there is no way to settle this issue except to observe for another session.

We should not be surprised if the third value is at least somewhat different from the other two. However, if the three values are not wildly different, they may begin to tell us something about responding in this phase. Still, it would be easy to admit that we do not yet have a very complete picture of what responding is like in this phase. After all, our participant has had only limited exposure to this condition, and we know it can take a bit of time for responding to adapt to a new set of influences. In other words, there is good reason to anticipate that the initial impact of our baseline condition may not be a ~ery good prediction of how responding might change with increasing experience.

As we keep collecting data from one session to the next, our graph will gradually draw an increasingly comprehensive picture of responding. With some luck, we may find that responding under this initial condition is relatively :able. This means t~t ~~sponding is neither generally increasing nor decrea~ ~ and th~t the variability from one value to another is not excessive and 15 i

fat~l~ consistent. We may even begin to feel some confidence in answering the ongmal f ·Wh · hi 5 .. ques ion. at kind of responding represents the typical impact oft cond1tton?

193

THE STEADY-STATE STRATEGY

BOX 9.1

Measuring One Participant M . Many Participants Once any Tunes versus

There are two different approach .. · al es to obta1111ng . experunent conditions on respond· U a picture of the effects of

di . th h mg. nder both cont 1 d . tions, e researc er can measure th b h . ro an treatment con- . · e e avior of one p rt· · or many participants once. Although . a icipant many times

there is a big difference between t:ou c: wmd up with lots of data either way; things about behavior. ese ternatives for our ability to discove;

To understand this argument rem b h . ' em er t e disc · · fact that behavior is a biological ph uss10n m chapter 2 about the

enomenon This h dearly observed only at the level of th . d' . · means t at behavior can be influence of any variables on behav· e m iVIb dual organism. In other words, the

. tor can e clearly n1 . the behav10r of each participant. Althou h . seen ~ Y as they impact variables would affect different partici g. we nught wish that treatment assume that thi ill b h pants m exactly the same way; we cannot

. s w e t e case. In fact, this is part of what we are trying to learn and usmthis~ group~d data from many different individuals makes it difficult t; answer quest10n.

As this chapte~ ~how~, observing the behavior of a single participant repeat­ edly under~ condition gives the researcher the opportunity to obtain a complete and clear picture ~f the effects of that condition on responding. It should be easy to se~ that ob~ervmg the behavior of a participant only once cannot provide the same inform:itt~~. It may not be so obvious that measuring the behavior of a large number of mdividuals once for each does not improve the completeness or clarity of the picture of the effects of the condition on behavior. Although this tactic would provide many observations, each would show only the smallest sample of how different participants, each with his or her unique characteristics and histories, might react to the condition. We would know no more about the effects of the condition on the behavior of each participant than we would if we measured a single individual once. In other words, the point is not merely to get a lot of data but to get enough of the right kind of data. What we need are data that reveal exactly how each participant's behavior is influenced by a condition.

Comparing States of Responding

With this answer, we may decide that we are ready to see how responding changes when our participant encounters an intervention condition. In order to make this comparison, we must first determine what kind of responding is typical of the effects of this new condition. As we accumulate data points across repeated sessions, the graph will gradually reveal a new picture of responding. we might find that responding is initially like that observed in ~he first phase but gradually transitions to a differe?t level. On the other hand, it might be that responding immediately changes m some way. Whatever the

9. STEADY STATES AND TRANsn10Ns 194

uld probably find that the more sessions we observe th . 'tial changes, we wo . di . , e 1111 ' d t d the effects of the intervention con t10n. better we un ers an fir h b 1·

. . art' cipant repeated exposure st to t e ase me condir By g1vmg our p 1 . hi ton d t1 to the intervention condition, we are trying to get a grap cal Picture

an 1 en di der each condition that is complete and representative Th

of respon ng un b nfid . at . . t t make sure we obtain enough data to e co ent that we hav 1s we wan o h d' · aft · e Ie~ned what responding looks like unde~ ea~ con 1t10n er its effects on the target behavior are fully developed. This will allow us to ~ompare respond­ . under the two conditions and be sure we are comparing data that fully mg Thi · · rt t b represent the impact of each condition. s ts tmpo_ an ecause we Want to be able to conclude that any differences in responding we see are due to the differences between the two conditions themselves.

The Risk of Extraneous Variables

What complicates this conclusion is the risk that responding under either condition might be influenced not just by the condition itself, but by extrane­ ous factors. Making repeated observations of a participant's responding under each condition provides one way of assessing this risk. This approach depends on trying to make sure there is relatively little variation from session to session in the key features that define each condition. This certainly does not mean that a participant's experiences are identical across sessions within each of these two phases. However, it does mean that the key factors that make up each condition are relatively consistent from session to session.

Given this consistency, if the data show that responding is unstable under either condition, we should assume that there are factors responsible for these variations in responding. We might reason that if these variations are not due to changes in the condition, which is being held constant, they must be due to extraneous factors. More optimistically, if the data are relatively stable during a condition we might assume that either extraneous factors are not having noticeable effects or that any extraneous effects are at least consistent.

In other words, when the data within a phase are relatively stable it provides limited assurance that extraneous influences are relatively minor. As we shall see, this conclusion is not necessarily true. However, it is at least reassuring that the data are not noticeably or systematically variable. This would leave us no choice but to worry that extraneous factors may be causing this variability. Uns~ble patterns of responding from session to session in a phase would req~~ us to admit that the data may represent not just the effects of the condition, but the effects of the extraneous factors as well. If this were the case, we woul~ ~heref?re not be in a good position to compare responding ~der that condition with responding under another condition. Such a comparison wo~d not allo_w us to conclude that differences in responding were due onlY to differences m the conditions. .

In other words, a graphical picture of stable responding across sessions under a c d't' be on 1 ion provides some encouragement that the data represent t

fJ-lE STEADY-STATE STRATEGY

-'Jects of that condition and th 195 o. l at the c . rniniIDal or at east constant. Stable 0 ntribution of e j(ind of marker for two important ;esponding-a stead xtrane~us factors is

ssed, stable responding sugg c 1aracteristics of tlley 1state- 1s therefore a

Cu ests th t c ata Fir t unJess they are consistent through a extraneous influ · s , as just dis-

out the c di . ences are mi · 1 suggests that any transition from the . . . on tton. Second, stabl ~a , enduring effects is complete. 1111ttal effects of the cond't' e res~ondmg

i ton to its more

summary

The steady-state strategy is an . th . approach to kin 1sons at mvolves ·measuring resp d' ma g experimental compar- under both control and experimental on md·g· for each participant repeatedly

con itions in . to assess an d manage extraneous infl succession. The objective is

uences and the b b . of respon din g that represents the full ft; re Yo tam a stable pattern evolved in the work of B F. Skinn de hi~cts of each condition. This strategy

· · er an s st d · first described in detail by Sidman

O 9

60 ) h u ents (Skinner, 1956), and was

1 way of managing extraneous influ · t as been a powerful and effective

ences and obtainin . 1 effects of each condition. This outcome all g a _c ear picture of the under control and intervention conditions ;:: ~ompansons of r~sponding

~~~ ::tr treatment variables. This focus :i = i:ili~~!~~c:::7i: . . e Y to hold up when tested by other researchers or used by

practitioners.

The steady-state strategy is equally useful in basic and applied research projects. Although it can be more challenging to obtain stable responding in nonlaboratory settings, the costs of failing to do so are unavoidable. If researchers collect only a few observations under a condition, the data cannot help to identify extraneous influences and will not provide a full picture of the effects of that condition. This limitation increases the risk that comparisons of responding under control and treatment conditions will be misleading. As a result, when others use this information, there is a greater chance they will not get the same results.

Practitioners often have the opportunity to measure a client's behavior repeatedly over a number of days or even weeks. A baseline or pretreatment phase is typically followed by an initial intervention designed to change responding in some practical way. The initial treatment procedure is often followed by adjustments required to make the procedure more effective.

9, STEADY STATES AND l'RANS)1'JONs 19 6

. also be needed to accommodate changes in the behavior or AdJustments ~ay s the intervention proceeds. Repeated measUr

uncling crrcumstances a . h e- surro h h allow practitioners to momtor c anges in th ments throughout eac P ase ~ . e

h · t11e project continues. target be avior as . interest of practitioners is delivering effective se

Of course, tl1~ pnmaryrun· ental comparisons for the purpose of publishinr­ vices not arrangmg expe d · f£ t t d · · g rese~ch findings. This obligation usually iscodurah ges e or hs o istmguish

f·~ t f …. eatment procedures an t e many ot er events going W.' between tl 1e e .iec so h . . . Ii d settings Decisions about when to make c anges m conditions are

on m adripp e b · clini'cal considerations than by steady-state criteria. As a often ven more y . . Service delivery priorities, practitioners are not usually in ul f th res t o ese h h . Ii ' b

a strong pos1 ·u·on to be confident about exactly w Y t err c ent s ehavior changes. This is simply one of the distinctions between research and practice.

STEADY STATES

Definition

A steady (stable) state of responding may be defined as a pattern of respond­ ing that shows relatively little variation in its measured dimensional quantities over some period of time. Let us examine some implications of this definition.

I Stea~y s_~e. A pattem~t r~~~wia]l~tf s~~r,~1~iiv~; li~-;:·1 ! vanat,or., m ,ts m~asured d1me.11~19p~Lq~~!J!t!l~~i-£?~~r;.:~or.n..~ per~od ' .. 1

0 Lf time· .. _ ·~-~:. ,_~_;-· _ ~~~-:_> ~· ;~L;;<::~S2}l1~bt1~i~.~~s~t~~t~;:~~;¢IU1-~~~~~-J First, it is important to understand that the meaning of the term steady is

relative. Exactly how much variability or what patterns of variability are required to describe responding as stable or unstable will vary from one research project to another. Such factors as the characteristics of the response class, the features of the general environment, the details of experimental pro­ cedures, and the focus of the experimental question may influence the researcher's judgment. For example, an investigator conducting a laboratory study of the effects of toxic chemicals on the behavior of rats may expect a very low level of variability, given similar studies already published. On the ot~er hand, a researcher conducting a project in a group home for individuals with mental retardation may need to accommodate the effects of day-to-day variations in extraneous factors that cannot be easily managed. Such dif­ ferences from one study to another mean that it is usually not feasible to define steady-state responding with a rigid formula .

. Second, although the dimensional quantity being measured might be stable, this does not mean that other quantities not being measured are also stable. F~r example, although the number of responses from session to sessi~n might show good stability, their duration could be systematically changing i1l

, I cADY sTA'fES

5'fw

n1e way. In fact, when two or m 197 soJ…- ore quanrt·

conunon for them to vary in dif~ 1 1es are being tU1 h . . 1.erent way measured 1 ·t. O lrnOW about t e stability of quant·t· S. f course th . , 1S not iv• • 1 1es that , ere 1s no . whY evidence of stability in a d" are not being way to 5 to the generalimensional quantity shoufdeasured. This ~e researcher make

. . . statement h not prompt :cnstead, 1t 1s more appropnate to say th at a particula t at respondin g ts . stable stable. r feature of respond· ..

'fhird, just because some aspect of mg ts respondin · to conclude that the environment i's al g is stable, it may not b . so stable A . e correct 1 of responding can result from a mix of h . · re at1vely consistent patt c angmg v . bl ern stable respon din g. Some environment 1 f: actors m aria es whose net effect · a ts ways but not inf] uence the target beh . ay even change in obvious . av1or. For exa 1 . wo seem to e an rmportant chan . mp e, a substitute teacher b uld ever, this change may not be evident ~e ~ a classroom environment. How­ experimental environment from ob Ill .t e data. All we can say about the extraneous environmental changes a:rvi_ng stable responding is that any have effects that are balanced by oth ect~g responding are either weak or

er environmental factors.

Uses

Evalu~tes ~easurement Decisions. The steady-state strategy is valuable because it . gutdes the researcher's decisions as the study progresses. This benefit begms by helping the researcher to evaluate prior decisions about how the target behavior is defined and measured. Collecting data under such rules for a number of sessions provides a picture of responding that may prompt second guesses about how the response class is defined, which dimensional quantities are measured, and how and when observation is conducted.

For example, the pattern of variability across sessions might suggest reconsidering the response class definition. If the data tend to fall into distinct patterns from one session to another (such as higher versus lower values), it could mean that the definition has combined different functional classes. For instance a definition of "aggressive" behavior may include both hitting and cursing. In some sessions, the target responses may be largely in the form of hitting, and on other days measured responses ~ay be mo_st~y cursing. If cursing rail t nds to occur at higher frequenaes than hitting, the data

gene ye h h Th … · ns -ri'th higher values t an ot ers. at 1s, sessions m s ow some sess10 – . Could h -.ainly in the form of cursing would have higher values whi h d'

c respon mg was hi u… h was 1 hi ttmg. · Thi pattern f . . responding most y s o th an . sessions . U1 w c t d ata are suggestmg v . . t the researcher to wonder if h e · a ari.ability llll;ght proi:e target behavior is defined. Perhaps it would be more

problem with hoW e hitting and cursing separately. useful to define and measurasures under each condition can also encourage

Collecting repeated me urement decisions. For example, if the data showed curiosity about other meas ne session to another, it might be tempting to very little change frc:,1 °was stable. We have already pointed out, however,

conclude that respon wg

9, STEADY STAT.E~ ANU TllANSITJONs

198 . de assurance that the dimension quantity be·

t ble data o n1 Y pro Vl fi . ltlg that s a . bl S h data may not reveal the reason or this stabiU

sured 1s sta e. uc . · · h ty mea . f hat is happening dunng sessions m1g t show th · T-~ al observation o w . . . at iu.1orm b h . var1· es a good bit from session to session m other Way the target e av1or . Id s.

ether with the overly stable data, these observatt~ns cou suggest t~at the Tog dure is insensitive to changes m the target behavior fio measurement proce fi hi . r

Solution is to address the reasons or t s msensitivity If some reason. 0 ne b · f · that Observation periods were too ne or not scheduled the pro bl em was . .

. t 1 they could be adJ'usted. Another solution is to measure other appropna e y, inti · · . . t't'es which might provide a more ormative picture of what d1IDens1on quan 1 1 ,

is happening with the behavior. . . As an example of these situations, consider data from partial mterval record-

ing using 5-minute intervals that showed co~sistently hi~ percentages of scored intervals. It could be that the relatively long mtervals result in most being scored as containing at least some of the target behavior. Interval recording procedures do not measure dimensional quantities, however, so the researcher might worry that there is interesting variation being missed in quantities such as count, duration, or frequency. Again, steady-state data might not always mean that all aspects of responding are stable. Such data must be examined in light of what is being measured and how it is being observed.

Reveals the Influence of Conditions. The steady-state strategy is espe­ cially valuable in revealing what is going on with the target behavior as it accumulates contact with a condition. It is typical that when a behavior is exposed to a new condition the behavior changes in some way. Although the new condition may have some immediate impact, the still recent experience with the previous condition may still be having some effect. In other words, the data often show a mixture of influences at the beginning of a new phase that reflects a transition in control from the previous condition to the new condition. Although this transition is sometimes a particular focus in some research projects, more often it is merely a nuisanc;:e because it complicates seeing a clear picture of the effects of the new condition alone.

As. t~e data show an end to the transition in responding that started when conditions changed, it is tempting to assume that the full effects of the new con~ition are finally evident. The data may now represent a level of responding that is ~otably higher or lower than in the previous condition. Although this ~hange m the level of responding may show the impact of the current condi­ ~100.' the steady-state strategy asks for evidence that the new level of respond- mg 1s durable That · th h 1 · is, e researc er needs to be sure that the apparently stab e responding will t ·

con mue as long as the condition is in effect. If responding were to eventually ch · . .

ange 1n some way, it would be important to include this h c ange as part of the ef£ t f h . . . graduall d ec s

O t e condition. For instance responding nught Y ecrease when an · t · . ' . d

exposure, this low level m en:-entio~ ts started. However, with cont1.11ue level that existed in the of ;;5pondmg ~t.ght gradually climb back to the higher capture all of th h P vious condition. The steady-state strategy helps to

ese c anges that might be characteristic of the condition.

oY srAtES s'l'Et

evaluates Experimental Cont l. 199 P ct· · ro Measu ·

r the same con 1tton can also al nng a behav1· uJl de ert the · . or repeated! us variables. Remember that any f: mvesttgator to the r 1 f Y oe O . actors th t o e o extra-

iOdependent vanable are extraneous to ex e/ are n~t explicitly part of the described, such extraneous factors may hp imental mterests. As chapt 8

. d e unrelat d er preparation an may occur unsystematically a fir ~ . to the experimental ever, they may also b~ at~ached to the general ~ircu e drill m a preschool). How- therefore have contmumg or systemat· f mstances of a study and m . ic e fects (inti ay for a study con d ucted m the workplace) Th uence from coworkers independent variable itself and there£ · ey can even be attached to the

. ore come and g . . withdrawn. (In struct1ons associated w·th O as 1t 1s presented and 1 example of this last category.) treatment procedures are a good

The steady-state strategy creates a g d of unsystematic extraneous factors th ~o .;::Portunity to detect the influence As we will see, instability in the data ca ~ t o~cur at some point in a phase. its sources must usually be guessed fan e rbelatt~ely easy to identify, although

. . rom o serv10g wh t · · sessions. It 1s usually more challenging t . d . a is go10g on during

O 1 factors that are consistent throughout a e~t~ the infl_u~nce of extraneous

0 1 1 well, their contribution may be missed ~ ~ ~o_n. If their unpact is stable as session to session however chan . · e~ unpact ebbs and flows from

' , ges 10 respond10g may hint t h · under an otherwise stable set of conditions. a t eir presence

The ste~dy-~tate strategy can help to identify unstable responding, but the real q~~st10~ ts what the researcher is able to do about excessive variability ?nee tt 1s evident. Studies differ from one another in how carefully extraneous influences must be managed. Some experimental questions and procedures require a high level of control, perhaps even a laboratory setting. However, even studies conducted in messy, real-world settings often require some management of extraneous influences. Whatever the requirements of an individual study, the level of stability in responding reflected in the data is a measure of the level of experimental control that the investigator has achieved.

Facilitates Experimental Comparisons. As we will see in more detail in the upcoming chapters on experimental design, the steady-state strategy provides the foundation for making comparisons between the effects of con­ trol and intervention conditions. Drawing conclusions about the effects of an intervention condition that have a good chance of being true, and therefore dependable, depends on how well the effects of both control and intervention conditions can be described and understood.

Efforts to establish stable responding under each condition help the investi- gator to do this in rwo key ways. First, repeatedly measuring responding under each condition helps to identify both the initial and final patterns of respond­ ing · h hase. second, these data also encourage efforts to manage

10 eac p h l . . . h . infl d h b ·ables which e ps to rmrunuze t e1r uence an t ere y raneous van , .. ext '&'. cts of each condition. These two outcomes of the steady-state

c I ar ifi es t h e e f.ie d" . . h b f't:. f h d' . h investigator to 1stmgrus etween e .1ects o t e con 1t1ons 1 e f other factors. The resulting comparison is therefore more strategy h~&'.p t

and the ef:iects o

9. STEADY STATES AND TRANsrnoNs 200

likely to be an accurate description of the effects of t~e intervention. 'fhis also means that the .findings have a good chance of holdmg up when others Use them in some way.

As an example, let us suppose that a researcher is conducting a study in developmental center looking at t~e _P?ssibili~ that a certain p~chotropi~ drug may make it more difficult for mdividuals with mental retardation to learn new skills. Each individual's performance, ~easured rep~atedl~ under control and drug conditions, will be partly a function of the basic testmg procedures used throughout both conditions. In addition, performance under the experi­ mental condition should reflect any influence of the drug. However, what if there are events going on in the participant's daily living conditions that vary from one day to another and that might affect their ~~rform~nce in daily testing sessions? One individual may be moved from one livmg urut to another a kind of disruption that often has broad effects on behavior. Another may g~ home on some weekends and behave differently on Mondays as a result. Still another may have health problems and may not be feeling well on some days.

Any effects of these extraneous factors may show up as variations in acquisition performance from one session to another. If the researcher ignores these variations and concludes that the difference in learning performance between the two conditions is due solely to the drug, this finding may not hold up very well for other researchers or practitioners. On the other hand, if the variations in responding within each condition are used to identify the contribution of the extraneous factors so that they can be better managed, it will be easier to identify the influence of the drug.

Identification

Trends. One of the challenges of the steady-state strategy is recognizing ':hen the data show stable responding and when they do not. There are par­ ticular features of variability in a series of data points that often bear on this decisi~n. One pa~tern of variability is called a trend. A trend may be defined as a relatively consistent change in a single direction. Although there are some excepti?ns, steady-state data do not show strong or consistent trends in either a decreasmg or an increasing direction.

It is not necessarily easy t O t 11 if h d . . d e t e ata m a phase are generally trending up~ar or downward. There are endless ways in which a sequence of data

~~::~;t:s s~w tren~s. The graphed data sets in Figure 9.1 shows some . · e data m both Panels A and B show a slightly increasing but

consistent trend How h · ever, t e greater range of the values in Panel B might

mas k th e f: act that the slop f h . · O Panel A. . e t e trend m these two data sets is the same as i.11

201 STEADY STATES

BOX 9.2

One Data Point at a Time

A trick that can help you apprec· t ti . ia e 1e imp t repeatedly under each condition . or ance of measuring respond" . . is to look at th d . mg

a a tune Pia . 1e mvesttgator sees them-one value t . e ata m the same way ti the first few data points on a graph Th · ce a piece of paper over all but

1 . . · en slowly sl'd h uncovermg successive values one at t· e t e paper to the right . a une. ,

If you do tlus for the graphs in Fi 9 1 . decide when the data show a trend.~ p· ' you will se~ …