ae ERGONOMICS, 1978, VOL. 21, NO. 8, 627-635 The Influence of Length and Frequency of Training Session on the Rate of Learning to Type By A. D. BADDELEY Medical Research Council Applied Psychology Unit, 15 Chaucer Road, Cambridge CB2 2EF and D. J. A. LONGMAN Post Office numeric code material using a Four groups of postmen were trained to type alphag for one or two hours conventional typewriter keyboard. Training was based on sessionslastin group given one session of occurring once or twice per day. Learning was mostefficient in the . Retention was one hourperday, andleastefficient in the group trained for two 2-hoursessions 30%. Again the about of speed in loss a ated ,andindic months nine or tested after one, three t is suggested that poorly.I most med perfor hours two of s session daily two for trained group uted over time rather than where operationally feasible, keyboard training should be distrib massed, 1. Introduction It has long been claimed that shorttraining sessions(distributed practice) produce faster learning than the equivalent amountof time spent with longer sessions (massed practice). Woodworth (1938) cites many studies which appear to support this claim across a wide rangeof tasks from archery to maze learning in the dancing mouse, and including the skill of learning to type. However, closer examination of the two typing Studies cited by Woodworth showsthat one is based on an experiment in which the groups compared differed in initial ability, making subsequent comparisonsinvalid, while the second turned outto be no morethan a statementthatthis aspect of learning to type should be studied. In recent years the tendency has been to assumethat the effect of distribution of practice is primarily a laboratory phenomenon occurring only underclosely specified conditions using simple laboratory tasks, and that even under these circumstances Its effects may be transient. A review ofthe literature (Bilodeau and Bilodeau 1961, p. 263) makes this point and remarks that “An aggravation to anyone who hasvaried distribution of practice on a standardpieceof (laboratory) hardware is the knowledge that somewhere, someoneis using his findings to urge an innocent consultee 10 | distribute the practice of his trainees as widely as possible”. 1976 Chapter 2) this apparent As has been pointed out elsewhere (Baddeley buted at least in inconsistencyin results of studies of distributed practice may be attri ng practice. The part to a failure to distinguish between two separate ways of distributi h early first of these is concerned with amountofpractice per day, andis a variable whic Studies showedto be very powerful(e.g. Perkins 1914), The secondis the length of the ful interval between successivetrials. Perkins (1914) foundthis to be a muchless power sively on this effect. None-the-less subsequent research has concentrated almost exclu of the interrole variable, producing a large volumeof inconclusive literature on the trial interval in verbal learning (e.g. Underwood, Ekstrand and Keppel 1964) and last 20-30 y, the pursuit-rotor performance (Bilodeau and Bilodeau 1961). Over the rs to have been potentially more powerful variable of amountof training per day appea i i B &: almost completely neglected. A. D. Baddeley and D. J. A. Longman 628 The question of optimal training schedules was raised again, however, Py theBr 1 8 Post Office, who estimated that with the adventof mechanical letter SOT Hs us se e machin prove necessary to train up to 10000 operators. Since the sorting standard typewriter keyboard,this in effect means teachin g 10000 po smn O ype clearly important that this should be doneas efficiently and economically asPo v*. It seemed likely that operators would be trained in their own sorting oO Hees, nm during training might be expected to carry on with their regular jobs, rather than | m full-time. With a relatively large number of men to be trained and a limited tra capacity, it becomes necessary to choose between either training operatorsintensive “ few at a time, or training a larger number of operators more gradually. The Paa the present experiment wasto provide information relevant to this decision by stuaying ; . the effect of distribution of practice on learning to type. Lo 2.5 did 80" sim pre (Ry Two lengths of training session (one or two hours) and two frequencies (1 tak or: separate groups of subjects were trained, one group receiving one session of one houra Ser sut dis rea sessions per day), were selected as being both operationally feasible and sulci } different to show any distribution of practice effects which might occur. Thus fou day (the | x 1 group), a secondgiven twosessions of one hour a day (the 2 x 1 group), a third given onesession of two hours a day (the 1 x 2 group), and the fourth given two sessions of two hours per day (the 2 x 2 group). 2. Procedure 2.1. General tra 3.1 For each condition, twelve postmen and six PHGs (Postman Higher Grade), ‘oh trained in groupsofsix. It was originally planned thatall groups should tram for “ Since testing was ona five-day week basis this meant training periods of twelve wee cac been hoped, training wasextended for a further 20h forall except the 1 x 1 group!© cor Wo rec completion -oftraining andtheinstallation of sorting equipmentitis of interestto kno en: how wellthis skill is retained, Subjects were therefore subsequently re-tested, either !, 3 rat for the | x 1 group, six weeks for the 1x 2and2x 1 groups and three weeks for the a group. However,since at the end of 60h levels of performance were not as high as fot which this proved impracticable due to previously arrangedleave. 1 Since in the operational situation there may sometimes be a delay betwee : » . or 9 months after completion o . . . W f training. g office. They were postmen © f d PHGs who claimed neither to have used a typewriter nor to be able to play a key instrument (pianists are said to be atypically adept at learni ng to type). Subjects W° tandomly assigned to the four groups, with the following contraints: (1) Each group contained 18 subjects, 12 postmen and 6 PHGs. (2) The groups were m atched on. the basis of age, the average for the four group 5 beingas follows: 1 x 1, 35-4 “9 1. Gy. Ages range from 19to 46.0 2.3. Equipment Six portable Imperial ty we Ter pre aff val 2.2. Subjects | All subjects were volunteers from the Croydon sortin | pri ee L S41 ys 1x2, 34:3 y; 2x 2, 34-6y. AB 60 Th the fac 3.2 3.2 all We Pri and po. © Specific features such as the representation of frequen tly tricts, Css * pewriters Ww ‘fed b assignive someof the peripheral keys by : . ere used. These had been modified used London postal dis Mc Rates of Learning to Type 629 2.4. Test and training material This comprised the alphanumeric post code material described by Conrad and Longman (1965) together with standard typewriter training material. . 2.5. Practice and test conditions Once subjects had mastered the keyboard, an attempt was madeto ensurethat they did one test per hour. Otherwise no attempt was madeto adhereto strict timetable governing rest pauses, tests etc., for the various groups. In each case the instructor simply tried to teach the group asefficiently as possible using the standard GPO procedure for typewriter training as described in the Post Office Training Rule Book (Rg 122) H. Since training was performed in conjunction with more normal duties, care was taken to ensurethat time oftraining session (a.m.or p.m.), and shift worked (early,late or regular day) were balanced across conditionsasfar as waspossible with the different training schedules. The two sessions in the 2x 1h and 2x 2h groups were always separated by at least 2h, one being in the morningandtheotherin the afternoon. One subject withdrew from the 1 x 1 h group, and data froma second1 x | subject had to be discarded since he was without his reading glasses for a month and hence could not read the test material adequately. 3.1. General 3. Results The simplest way of expressing the results would be to present a graph showing for each group the mean performance on each hour, averaged over all operators. In principle this will be done. There is however, the problem of days on which operators were, for some reason, absent e.g. because of a public holiday, or an illness. The first procedure adopted has been to treat successive working days as a continuous sequence, Thus a manin the 1 x 1 group for example, who wasnever absent, would yield performancedata for 60 h. A man whohad two days absence would have a tecord which terminated after 58h. This procedure doesgive a realistic indication of tate of learning, but has the disadvantagethatthe last few hours mayhave so few men represented thatthe dataareof little value. Furthermore, it might be argued that froma Practical point of view the possibility that one schedule wasparticularly liable to be allected by days missed due to absence, should be taken into accountin assessing the Value of the schedule. Forthis reasonallstatistical comparisonsare based ona nominal 60 or 80h of training, regardless of how manyofthese training hours were in factlost. Thusthe graphsgive an indication ofrate of learning per actual hour of training, while the tables andstatisticaltests show the rate of learning per nominalhour,ignoringthe fact that most subjects lost one or two hours during the training programme. 3.2 Performance after 60 hours oflists 32.1. §tage 1. Stage 1 refers to that period of training priorto the use . containing all codes, During this period operators could notbe said to be‘ operational ' since there Were still procedures and parts of the keyboard they had notyet adequately learned. Progress during this phaseis basically governed by the instructor who decides at what Point new material should be introduced. For administrative convenience It 1s almost “ssential that a class advances together andit is up to the instructor to decide at what Moment to move on. A. D. Baddeley and D, J. A. Longman 630 in} n the required totearsoak le 1 shows the mean numberof actual hoursof training whole keyboard under the four experimental conditions, together w1 values obtained. I a freq sign gro’ gro’ (p< . Schedule ixih 2xih 1x2h 2x2h Mean hours to learn 34-9 : 42-6 43-2 49-7 aavee“ 26-44 34-46 37-45 46-54 3.2. mo: pert The moststriking feature of this result is the slowness of the 2 x 2h sche dule. The i the I x 2or1 x 1, whichis easily fastest subject in this group is slower than the slowest in the fastest group. , "afte rela to ¢ wnin 3.2.2. Speed. Average speed of performanceontest runs for the ourero'spi . Figure 1, together with the equivalent data from the typewriter ey iy a few subjects Conrad and Longman (1965) study. Since observations based on on Y eon omitted can be very misleading, points with data from less than six operators ha especially As with Stage 1, the moststriking feature is the slowness of the 2 x 2 group,esp CORRECT KEYSTROKES /MIN. compared with the 1 x 1 group. 90} 8O} 7OF 6OF 50 40 50 60 HOURS OF PRACTICE i J. 70 Ane —_——_ 80 bet ie based Figure 1. Rateof acquisition ofty Ping skill as a function oftraining schedule. The 1°5 , x 2 func tion is base Sh per day. on Conrad and Longman’s (19 65) typewriter group who trained for two sessio of 1 ns A Statistical anal YSIS was perf [lowing 4 ormed using each subject’s speed onthetest e with the nominal 60h oftraining. The mean numberof actual Ixth 79-3} Range Mean actual 53-3-95-7 Range 45-60 hours Sigr Casi mean speed for the four conditio hoursoftraining are given in Table Table 2. Mean Rate (Correct Keystrokes/m in) after Mean Rate cert 35-2 Schedule 2x th 73-43 54-4-98-] 543 46-58 2. ns together a nominal 60h oftraining 1x2h 71-12 43-9-101-2 53:5 h an 166.823 8-9 <0 60 46-58 . with A nalysis of Variance showed a hi gh sessions of one hour ly Si gn if ic ef an fe t ct of le being bett er than sessions of two (p<0-01), ngth of sessionec0 and a smaller eff 3 det this Wit! oft ERG. Rates of Learning to Type | | 631 frequency of session, with one per day being better than two (p=0-05). There was no significant interaction. Comparison between groups, using a ¢ test, showed the 1 x Lh group to be faster after 60 h training than the 2 x 2h group (p<0-01) or the 1x 2h group (p< 0-05,1 tail), and the 2 x 1h groupto bereliably faster than the 2 x 2h group (p<0-05). 3.2.3. Errors. (i) Uncorrected Errors. From a practical point of view these are the most important type of error, since they maylead to a mis-sortif the error produces a permissible post code. Their occurrencein the four groupsis shown in Figure 2. Results after 60h for the four groups are shown in Table 3. Since error scores are based on relatively small numbers of error responses, data from thelastfive tests were used so as to give a more reliable measure. 2x! a—aA /x2 u——K 2x2 O a > id oO Nu wn r w O Ixfi Oo—o Figure 2. 40 50 60 HOURS OF PRACTICE 70 | | Oo wi Lal un UNCORRECTED ERRORS (°/, OF TOTAL KEYSTROKES) Schedules o—e 80 Mean rate of uncorrected errors as a function of training schedule. Analysis of Variance showed length of session, frequency of session and the interaction between length and frequency all to be highly significant (p< 0-001). This is almost certainly due to the very high error rate found in the 2x 2h group, which produced Significantly more uncorrected errors than any of the other groups (p<001 in each case). No other inter-group difference wassignificant. Table 3. Mean Percent Uncorrected Errors after a Nominal 60h of Training: Schedule Percent uncorrected error Range we, 2x2h —. Ixth 2xth tx2h- 1-09 1:14 1-41 2-06 0:22-2.18 0-06-2.42 0-40-3-45 038-465 (ii) Corrected Errors. In all conditions subjects madeerrors which were immediately detected and corrected, Such errors constituted about0:5% of keystrokes in all groups, this level remaining remarkably stable throughout training. This result is consistent with Rabbitt and Vyas’s (1970) suggestion that subjects use detected errors as a means of Monitoring Tor ERG. performance. oo oe : oo, 2Y _ A.D. Baddeley and D. J. A. Longman 632 3.4. Performance after 80 hours Since all but the 1 x 1 group continued up to a nominal 80h of training, a second analysis was performedat this point. Table 4 shows the mean keying rate at this stage, excluding four subjects who had missed 20 or more hours oftraining. Statistical analysis confirms the results of tests after 60h in showing a significant difference between the 2 x 1 and the 2 x 2 groups(p <0-05) but not between the 1 x 2 and the 2x2 groups. Table 4. Mean Rate (Keystrokes min~' after a nominal 80h of training. Schedule Mean Rate 2xih 1x2h 2x 2h 89-4 82:8 776 Range 68-5-119-8 A7-8-117°5 §5-4-93-2 Range 64-80 69-80 70-80 Mean Hours Actual Training 717-4 75-4 17 17 16 No.of Subjects 79-2 3.5. Retention Subjects from each group weredividedinto three sub-groups of approximately equal skill as measured by performance at the end of training. One such sub-group was retestedafter approximately 1 month,oneafter 3 months andthethirdafter 9 months. Retesting involveda session of one hourin which subjects first spent a few minutes reme secol give retes type It. of sp subs grou reas¢ 3.6. At they “very subji to th less 1 Seco: thes grou ‘warming-up’ by practising simple phrasesetc., followed by a 15 min timed test run 0? post code material. They were then given a number ofshort stimuli which aimed to Tf) 90t A.) B.: C.. Dz: x , 80} \ x x x <= = wl * wo wv oO . & ke & w 60F x . - 0 ns © Oo UO z x oe 2xl SOF . ta = K2x2 40} Sm , __ xd x 2» a S tin PO * xxl 33 2x2 = 30h, Wy eo 0 Figure 3. 1 Retention of typi r 3 RETENTION INTERV AL (MONTHS) Table 9 ng skill over a 9 month period as a function oftra ining schedule Rates of Learning to Type 633 remedy any obvious defects shown onthefirst run. In the last 15 min they performed a second test run. Figure 3 shows performanceon this second run, which seemslikely to give a truer indication oftheir retained skill than the earlier run. It proved possible to retest 62 of the 70 subjects who completed training. Of these 7 claimed to have used a typewriter since completing training andtheir retention data were therefore discarded. It is clear that someforgetting does occur, but the results seem to suggest thatthe loss of speed is not excessive (approximately 30° after 9 months), and does notincrease substantially after the first three months (with the possible exception of the 2x 2h group). Similarly, although error rate does increase, for most subjects it remains within reasonable limits. 3.6. Subjective Ratings. Atthe timeof retention testing, subjects were asked to rate the schedule on which they had trained as‘very satisfactory,’ ‘satisfactory’, ‘reasonable’, unsatisfactory’ or ‘very unsatisfactory’. The first section of Table 5 showstheresults of this rating. While subjects in general responded favourably, the trendis clearly in the opposite direction to that found in termsof learning efficiency, with the 1 x 1 h per day group producing a less enthusiastic response than the 2 x 2h per day group. Thisis also reflected in the second question regardingthe preferred schedule (see Table 5): subjects tendedto prefer theschedule they had experienced, but this is much more pronouncedin the 2x2h group than for the 1x 1h schedule. A similar pattern’ is- discernable on the third Table 5(a). Responses to the question “ Howsatisfactory did you find yourtraining schedule?” Very Schedule Satisfactory A. 1x1] h/day 7 2 1 B. 2x 1 h/day C. 1x 2h/day D. 2x 2h/day 1 6 7 7 Very Reasonable 6 4 5 3 4 3 2 0 3 3 Unsatisfactory Mean 2 2:40 5 2. 2 0 0 0 0 1:86 2:00 1-73 . Table 5(b). Responses to the question choose? you would schedule which again, “Tf trained Schedule Trained A A. 1x Lh/day 5 C. 1x 2h/day D. 2x 2h/day 0 0 Schedule C 1 D 6 3 4 3 6 1 B. 2x 1 h/day Chosen B 6 11 g 1 2 3 7 Responses to the question Table 5(c). “ How keen would you be to undergo further training on the same schedule? Be very . C. 1x 2h/day T.able 5, keen 1 i Take part 4 Refuse 5 Mean 2 if necessary 3 5 0 0 0 131 Soy: of responses i, ratings on the four schedules studied. Numbers represent the distribution Subjective to the three questions as a function oftraining schedule. Not all subjects could be contracted, an Some subjects failed to respond to one or more questions. A. D. Baddeley and D. J. A. Longman 634 question, regarding willingness to undergo further training. In general, ems seems to be high with the exception of two 1 x 1h subjects who would refuse further laid Twoconclusions may be drawn from this pattern of results. First, 1s seems unlikely grou deu> Q dure raph perf: that the better performanceofthe 1 x 1 h group is due to a higher level of mouryalion If anything they were less motivated and resented having to spend 12 weeks learning w at some of their colleagues were learning in 3 weeks. This should obviously be borne in mind in evaluating possible schedules,althoughit should not be given too much weight mas: since motivation remained high even in this group, and might have been even higherif subjects had not had colleagues simultaneously undergoing a more rapid training, kon Peri Tra! absc Ver zwe Ma: Secondly it is worth noting that an evaluation of the various schedules purely in terms of subjective ratings would have led to a recommendation of a schedule which 1s associated with the slowest learning, the least accurate performance and the poorest retention. As such it amply reinforced Poulton’s (1976) warningof the dangerof relying exclusively on subjective assessments. _ BAL Bric 243Co» 4, Discussion The decision as to what schedule should be used to train operators will clearly depend heavily on operational considerations. Nevertheless, the results of the present studyaresufficiently clear-cut to allow certain recommendations. In general, a session of 2h seems to be too longforefficient training. One hour appears to be a more satisfactory length. Onesession per dayis slightly moreeffective than two, though m is not a very greatdifference.It is clear that the 2x 2h training schedule is not a g00 PER 253Pou RAt one. Despite the fact that fewer hours were lost on this schedule due to absences, holidaysetc., it produced consistently poor performance in both speed and accuracy after 80h). It is interesting to note that the typewriter keyboard group in the Conra and Longman (1965) study showsa very similar rate of progress. This group was g!ve" {Wosessions of 1-Sh per day, separated by a half-hour break. ned Theretention tests indicate that keyboard training is reasonably well retain’ ; isnot Although both speed and accuracy deteriorate if the task is not practised, the loss 18 enormous. In general terms, this study supports those carried out in the early years off this Le ingle century in suggesting thatit is a mistaket o try to cram too muchtraining .; into a rat session, orindeed a single day. At a more theoretical level it suggests ed preoccupation with the small and unreliable inter-trial interval effect has distrac’ attention from the more robust and powerful effect of the limit on amountof learning that can be accomplished Sned per day. Such effects are likely to be of considerable practic importance; they raise many further questions including: (1) If two tasks are to earned concurrently, how similar must the second task be to impairthe rate oflearniné mach Is thentage effectoflimited motorsted kills as typing? inst ,ist (3) mh the (2) disadva massintog discrete can be accoun forsuch by intra-d ay chan t ‘uctor (or Instruction)effectiveness? Since there is at present no theory of lear Which predicts such effects they also represent a considerable theoret ical challeng gues, to Postmaster, Mr. W. H. Line, to the ChiefInspector, Mt” t and Practical help, and of course the local Tepresentatives of the UPW who gave both general to the m €n who Suppo! took part. . Uni pmentonsen ne (the 1 x 1 group wasfaster and more accurate after 60h training than the 2 x 2 grou Ma Rey Rates of Learning to Type 635 | code alpha-numérique a Quatre groupes d’employés des Postes ont ete entrainés a taper des données en cours de sé¢ances d’une au effectué ete enta aide d’un clavier de machinea écrire traditionnel. Cet entrainem le plus efficace pour le été issagea L’apprent jour. par fois deux ou une u durée de une ou deux heures ayantlie groupe travaillant en pourle cace groupe bénéficiant d’une seule séance d’une heure par jouret le moinseffi mois. La perte de la 9 ou trois un, aprés testée été a rétention deux scances de deux heures par jour. La avéré étre le moins — s’est heures’ deux de séances ‘deux groupe le encore La 30%. rapidité était d’environ temps, plutdt que le dans distribuée sage performant. On suggére qu’il vaut mieux instaurer un apprentls clavier. un sur frappe de ssage !’apprenti massé, du moins en ce qui concerne Vier Gruppen von Postangestellten wurden in der Eingabe eines alphanumerischen Codes in eine konventionelle Schreibmaschinentastatur eingeiibt. Die Einiibung erfolgte in ein- bis zweimal taglichen Perioden von ein bis zwei Stunden Dauer. Die Ubung war besser in der Gruppe, die eine einstiindige Trainingsperiode pro Taghatte, und schlechterfiir die Gruppe, die zweimal zweistiindige Trainingsperioden absoliverte. Die verbliebene Ubung wurde nach einem, drei und neun Monaten ermittelt und zeigte einen lich je Verlust an Arbeitsgeschwindigkeit von ungefahr 30%. Auch hier wiederum arbeitete die zweimaltag ein fiir wird ist, ar durchfiihrb es zwel Stunden eingesetzte Versuchsgruppe schlechter. Wo immer empfohlen. Uben massierten dem gegeniiber Uben Maschinentastaturtraining das verteilte References BappDELey, A. D., 1976. The Psychology of Memory. New York Basic BOOKS. ea ek E. A., and Br.oprau, I. M., 1961. Motor Skills Learning. Annual Review of Psychology, 12, Conrad,R., and LONGMAN, D. J. A., 1965. Standard typewriter versus chord keyboard-—an experimental comparison. Ergonomics, 8, 77-78. : ao FD ee N. L. 1914. The value of distributed repetitions in rote learning. British Journal of Psychology, 7, PouLTon, E. C., 1976. Quantitative subjective assessments are almost always biased, sometimes completely fh R 7? laboratory choice RT tasks. In A. F. Sanders (Ed.), Attention and Performance 111, 56-76. Amsterdam: NorTH HoLLanb. UNDERWOoD, B. J., EKsTRAND, B. R., and KEPPEL, G., 1964. Studies in distributed practice: XXIII. nd resetetinPmat s t tet TS 2a by fr. misleading. Bulletin of the British Psychological Society, 29, 385-387. ABBITT, P. M. A., and Vyas, S. M. 1970. An elementary preliminary taxonomy for some errors in w Variations in response-term interference. Journal of Experimental Psychology, 68, 201-212. oopWwoRTH, R. S., 1938 Experimental Psychology. (London: METHUEN). Manuscript received 1 February 1977 . evised manuscript received 11 July 1977 .