R.D. Nelson et al- A Linear Pendulum Experiment: Effects of Operator Intention on Damping Rate

8/3/2019 R.D. Nelson et al- A Linear Pendulum Experiment: Effects of Operator Intention on Damping Rate

1/19

Journal of Scientific Exploration, Vol. 8, No . 4, pp. 471-489, 1994 0892-3310/94 1994 Society fo r Scientific Exploration

A Linear Pendulum Experiment: Effects of OperatorIntention on Damping Rate

R. D. N E L S O N , G. J . B R A D I S H , R. G. J A H N , A ND B. J. D U N N EPrinceton Engineering Anomalies Research, C-131, School of Engineering/Applied SciencePrinceton University, Princeton, N J 08544

AbstractA n attract ive pen dulu m consist ing of a two -in ch crystal ball sus-pended on a fused silica rod is the focus of an experiment to measure possibleeffects of conscious intention on an analog physical system. The pe ndu lum isenclosed in a clear acrylic box, an d provided with a computer controlled m e-chanical system to release it from th e same starting height in repeated runs. Ahigh speed binary co un ter registers in terru ption s of photodiode beams, tomeasure velocities at the nadi r of the pe ndu lum arc with microsecond accura-cy . In r u n s of 100 swings, taking about three minutes, operators attempt tokeep swin gs high, i.e. to decrease th e da m ping rate (HI); to reduce swing am -plitude, i.e. to increase th e dam ping rate (LO); or to take an un disturbed base-l ine (BL) .Over a total of 1545 sets, generated by 42 operators, the HI - LOdifference iss ign i f ican t in the direction of in ten t ion for f ive individuals, and the differ-ence be tween in ten t ion an d baseline runs is s ign i f ican t an d positive for fiveother operators. The overall HI - LO difference is reduced to non - s ign i f i -cance by strong n egative performan ces from several operators, four of whomhave com parab ly large scores in the direction opposite to i n t e n t ion . A na lysi sof var ian ce reveals significant intern al structure in the database (m ain effectsF4, 189= 2.845, p = .025). Subset comparisons indicate that male operatorstend to score higher than fem ales, and that rand om ly instructed trials tend to-ward high er scores than vo litio n al trials, especially for male operators. T rialsgenerated with the operator in a remote location have a larger effect size thanthe local trials.W hi l e direct comparison s are not straightforward, it appears tha t effects ofoperator in ten t ion on the pen dulu m damping rate may be similar in magni-tude an d style to those in other human /machin e interaction experiments. A l-though this result fails to support an experimen tal hypothesis that th e analognature of the pen dulu m experimen t would engender larger effect sizes, it doesconfirm a basic similarity of consciousness effects across experiments usingfundamen ta l ly differen t physical systems.

IntroductionExperiments using electronic random event generators (REG's) of severaltypes (Nel son , D un ne , & Jahn , 1984; Jahn , D unn e , & Ne lson , 1987; Nelson et

471


2/19

472 R . D . Nelson e t a l .al., 1991), as well as a random mechan ica l cascade ( R M C ) exper iment ,(Dunne , Nelson, & Jahn , 1988) have provided evidence for anom alous corre-la t ions of the performance of such physica l devices with operator in ten t ion . Inparticular, shifts of the empirical distribution means have been found to be sig-nificantly correlated with volitionally or randomly assigned intentions to in-f luence them. A l though these expe r iment s are based on su bstantial ly differingphysical processes, they are all essen tially digital or discrete in nature , with bi-nary positive or negative increments in the experimental measures as the tar-get of the operator 's in tent ions. To increase its generality, this genre of re-search has been extended into the analog domain via an experiment that haspotentially greater sensit ivi ty to operator interaction because of its appealingaesthetic qualit ies and the cont inuously variable nature of the measurable.The device is a classical l inear pendulum, suitably instrumented to provideprecise measurement of its dynam ic performance and to give appropriate feed-back to operators. Of the many possible configurations, a free-swinging pen-dulum enclosed in a clear acrylic box was chosen for development , with thedamping rate selected as the primary m easurable. Volunteer operators, non eof whom claim special abilities, are directed to sit quietly about one and a halfmeters from the pendulum and focus attention on it with either a HI in ten t ion ,defined as keeping the swings high (corresponding to a decrease in the damp-in g rate), or a LO intent ion, defined as keeping th e swings lo w (correspondingto an increase in the damping rate ), or to take a baseline (B L), wherein there isno effort to change th e pendulum behavior. These HI, LO, and BL condit ionsare accumulated in con t iguous sets of three runs , which are then compoundedinto series which are considered to be independen t replications of the experi-ment .

EquipmentThe pendulum bob is a clear quar tz crystal ball tw o inches in diameter at-tached to a 30-inch long, clear fused silica rod, chosen for its extremely small

coefficient of thermal expansion. The upper end of the rod is mounted in brassan d aluminum fixtures holding bearing elements. In pilot studies, three typesof precision bearings were tested, in a search for optimum reliability and mini-mum bearing con tribution s to damping forces. The f inal choice for the formalexperiment was a miniature dual-race ball bearing system that is highly reli-able, with lo w friction and no detectable sensitivity to wear or thermal expan-sion. These bearings con tribute on ly a small fraction (about 7% at maximumarc) of the composite damping forces; the rest are presumably aerodynamic.The bear ing system is supported by a massive a lu m inu m bar, 2.5 inchessquare, which is f ixed to a vertically or iented, m achined a lu m inu m plate, 1.5inches thick, attached in turn to the aluminum baseplate of the experiment.T he entire pendulum assembly is enclosed in a clear acrylic box, 24 inchessquare by 36 inches tall, which stands on a massive support table that enclosesth e electronic hardware. Figure 1 is a photograph of the device from th e oper-


3/19

Linea r Pendulum Exper iment 473

Figure 1: The L inear Pendulum Experimental Deviceator's point of view, and Figure 2 is a close-up photograph showing the bob atth e beginning of a run .A stepper motor, mounted behind the vertical plate and operated by a micro-controller, moves an arm that pushes th e pendulum bob up to a start cradle on


4/19

474 R . D . Nelson et al.

Figure 2: The Primary Components of the Linear Pendulum

the r ight side of the case. In the photo, the bob has just been released from th ecradle by the mechanical arm, which may be seen accelerating ahead of thebob. Behind the bob is a conical shield over a thermistor, and at the far left isthe photodiode system. Figure 3 (Data Procesing) sketches the mechanical de-tails of the measurement system.To initiate a run , the arm is rapidly moved to its parkin g position on the op-posite side of the case, releasing th e pendulum to swing freely through an arcof about 35 degrees. A double blade is mou nted on the rod near th e bob, pro-t ruding toward th e backplate, so that the two leading edges, separated byabout one centimeter, pass through photodiode pairs mounted on a stalk at-tached to the backplate. The interruptions of the photodiode beams are timedwith 50 nan osecond resolution, using a binary counter with a 20 megahertzclock rate, and the times are recorded as raw data in computer files, togetherwith computed velocities an d changes of velocity (damping) , an d ident i fy ingindex information. It should be noted that although the damping rate is thespecified target of operator in tent ion, we canno t exclud e the possibility thatthe experimen tal results could reflect inf luences on other elements of the sys-tem, such as the measuring circuit, which includes both digital and analogcomponen ts .Direct feedback to the operator during the run is provided by light projectedalong th e fused silica rod to the crystal bob, th e color of which is chang ed by agraduated f i l ter whose posi t ion is controlled by the magn itude and s ign of cu-mulated differences be tween the ongoing run and the preceding baselin e ru n .


5/19

Linear Pendulum Experiment 475T he operation of the exper imen t is ent i re ly controlled by co m p u te r soft-ware tha t communica te s with the physica l appara tus th rough a GPIB IEEE-488 interface and EPR OM based micro-controller code. The program is writ-ten in G W B A S I C to run on an 80286-based PC; it manages th e experiment ,including all operator activity, an d mainta ins data files with their correspond-in g index. For data security and integri ty, a r edundan t hardcopy of blocks ofaveraged raw data is made to provide confirmation of the primary data an dprotection against it s loss. In addition to the hardcopy, a com plete copy of thedata is writ ten to a floppy disk as well as to the hard disk. A set of automatic in -s t ruments (Sensor I n s t rumen ts Co., Inc.) records tempera ture , humidity , andbarom etric pressure, an d these parameters are inc luded in the index for eachr u n .The massive pendulum structure itself is level an d stable relative to the flooran d building, and while building vibration or the effects of passing traffic, etc.,t ransmitted through th e con crete slab floor may in principle affect data at a sta-tistically detectable level, the experimental design ensures that such effectswill not be correlated with condi t ions of in tent ion or any secondary parame-ters. Physical movements of the operators, such as swaying, tapping, rocking,an d head noddin g in response to the pendulum, are a potential in f luence if theyshould be mechanically coupled to the pendulum via the floor or air move-ment . The latter possibility is largely obviated by the complete enclosure, butprotection from mechanical interference, such as stamping on the floor ortouching th e pendulum case, or acoustical disturbance such as shouting orwhistling, is currently provided only by operator training an d integrity. How-ever, testing indicates that th e most prominen t effect, an d indeed th e only de-tectable change due to regular, synchronized mechanical interference (e.g.,tapping on the case), is to in crease the var iance of data with in run s , leading toincreased standard error and hence more conservat ive tests of differences be-tween runs . Co m plete protec t ion f rom all such spurious sources of effect is in-herent in the subset of the database where operators are in a remote locationduring th e run . A ll of these are run automatically, often when no one is in thelaboratory, and in any case the staff do not know the order of the operator in -tention s, so that there is no possibility for conscious or un conscious in t roduc-tion of correlated vibra tions . Finally, a fail-safe threshold check is incorporat-ed to detect r ampan t outliers caused by bearing malfunction or other majorartifact, but in the formal database accumulated since the installation of theprecision dual-race bearings, n o such threshold even ts have occurred.

ProcedureD ata are taken in runs of 100 full swings and the pendulum period is about1.8 seconds; a run thus takes about three minutes, plus time for writing filesan d recording summary in format ion. T he three in ten t ions are combined incontiguous sets of three runs , wherein th e envi ronmenta l an d mechanical con-ditions are presumed to remain closely similar. Following a non-recorded run


6/19

476 R. D . Nelson et al.that verifies nominal system performance, data are generated in sessions thatlast about an hour.Feedback during a session is based on comparison of the h i g h and low inten -tion runs w i t h the baseline of the current set. This requires that the f i r s t run inall sets be the baseline, an d hence only the comparison of randomly orderedhigh versus low runs is s t r i c t l y immune to secular trends in the machine's per-formance. A c t u a l l y , no such trends have been seen in extensive calibrationtests, with the exception of some correlations w i t h changes in atmosphericvariables, an d these variations are typically much too slow to have an y bearingon within-set comparisons. More specifically, temperature, humidity, andbarometric pressure are r o u t i n e l y recorded, and the calibration data indicatethat barometric pressure is correlated (r = -0.93) with the overall change in ve-locity during a run. Temperature and h u m i d i t y are also related, but cross-corre-lated w i t h pressure, so that a regression model using only pressure can accountf o r about 90% of the variance f r o m these sources. Analysis s o f t w a r e calculatesa correction factor from barometric pressure readings to compensate for the in-f l u e n c e of the environmental variables on the pendulum damping rate, but be-cause comparison s are all made within the sets, this correction is negligible.The computer program includes an option for delayed start of a sequence ofpreprogrammed runs. This is used for overnight calibrations, and also for f o r -mal experiments with the operator in a remote location. In the latter case, anarrangement is made to generate data in session-length blocks beginning at aspecified time, w i t h runs spaced at f i v e minute intervals. The operator reportsthe order of the HI and LO intentions a f t e r the data have been generated andrecorded, and o n l y then receives feedback.The experimental parameters maintained in the index and logbook includethe mode of instruction and the mode of feedback. Operators may choose theorder of HI and LO intentions, or have the order assigned randomly by the pro-g ram . There are several options for feedback, i nc lud ing digital or color indica-tors, or both, or the operator may choose to have no explicit feedback. The dig-ital option is a computer display of the positive or negative cumulativedeviation of the present run's change in velocity in each half swing, comparedw i t h the baseline. Color feedback shows increasing positive deviations asamber, then red illumination of the pendulum bob, and negative deviations asgreen, changing to blue. Operators are encouraged to generate multiple series,and to explore the optional instruction and feedback modes. The plannedanalyses include comparison of these options as well as comparison of indi-vidual operators and the male and female subsets.

Data ProcessingExtensive processing is needed to transform the original data stream intowell-behaved random variates, suitable for statistical comparisons to deter-mine operator e f f e c t s . The ultimate goal is to reduce the measured data for

each run to quant i t ies tha t may be compared using robust parametric tests. T he


7/19

Linear Pendulum Experiment 477

Figure 3: Photodiode Measuremen t Sys tem

original data are auto correlated, nonlinearly decreasing values, and are con-verted to normalized scores that are more nearly linear and are uncorrelated.The final comparisons are based on subtractions between nominal ly equiva-lent points in the normalized time series, compounded across th e run. The re-sulting summary numbers are well-behaved, as described in more detail in theCalibrations section.A complete description of the analysis including details of the normaliza-tion is provided in Nelson & Bradish, 1993. Briefly, the measurement/analysislogic proceeds as follows: Two photodiode pairs mounted at the nadir of thependulum swing have their respective light beams interrupted by the doubleblade on the pendulum shaft (Figure 3). Blade edges a1 and a2 interrupt photo-diode beam A as the pendulum swings rightward; times A1 and A 2 are readfrom a 50 nanosecond resolution clock (32 bit binary coun ter) and recorded incomputer memory. Similarly, passage of edges b1 and b2 over detector B on theleftward swing are recorded as times B1 and B2For each such half swing, the raw data are thus tw o interrupt times that arerecorded along with a status byte that identifies left an d right swings and achecksum that validates data transmission. From these times an d the distancebetween edges, corresponding right an d left velocities, V, ar e calculated. Foreach run , an average change in velocity from swing to swing, V - i -Vi+1, is com-puted and normalized by the current swing velocity. For each half-swing thenormalized change in velocity is:

A grand mean for the run is computed across all half-swings, an d differencesof the means for the HI and LO intentions are assessed using Student t-tests;fo r convenience, these scores may be converted to standard Z-scores, i.e. ex-


8/19

478 R . D . Nelson et al.pressed in units of the standard deviation of the mean , via the inverse normaldistribution. These calculations are made for each series, and the results pro-vided to operators as feedback. For concatenat ions of more than one series,analogous computations are made by compounding the individual tripolar setswithout regard to their original series membership. In addition to the H I - LOcomparison, an orthogonal comparison is made of the combined in tent ionalrun s vs baselines (INT - BL) . For the pre-planned comparisons, a directionalprediction w as made, and the 0.05 criterion for "significant" deviation corre-sponds to a Z-score of 1.645.

CalibrationsSince theoretical modeling of ideal pendulum function provides only a

rough approximation to the precise em pirical measures of a complex, real sys-tem, experimental data can only be assessed against a background of calibra-t ions that characterize the performance of the pendu lum in the absence of op-erator interactions. For calibration runs, the computer program provides fullyautomatic con trol of the machin e and permits delayed start times so that datamay be taken during th e night when there is little building activity and no peo-ple present, as wel l as during normal laboratory hours. Calibrations were doneas sets of 27 runs , some taken in single sessions typically beginn ing at 2:00am, and others during the day, to determine whether the activity of people inthe laboratory could detectably inf luence pendulum performance. These tw ocategories of calibrations ar e indistinguishable. To assess distribution charac-teristics an d confirm th e validity of the statistical processing, th e calibrationdata were arbitrarily assigned to the three intention categories, then processedas if they were experimental data taken in 9-set series, with comparisons madeof th e "HI" an d "LO", an d both of these with th e "BL". This rando m assign-ment procedure was used to construct 600 artificial series t-scores. A good-ness-of-fi t comparison of these with th e appropriate theoretical Studen t t-score distribution yields x2 = 11.964, based on 13 df, with a correspondingprobability of 0.531. Thus , although some session-to-session changes due toatmospheric effects are detectable in the mean and standard deviation, theseare no rmalized correctly by the within-set differential analysis.

ResultsThe formal experiment began on Jan uary 10, 1990; on February 1, 1993, thedecision was made to conclude th e global accumulation of data, and thereafterto limit data collection to the production of large individual operator databasesfor systematic exploration of secondary parameters. This report summarizes

th e primary results of a comprehensive analysis of the three-year pendulumdatabase.The entire formal database contains 235 complete series and 5 partial series,for a total of 1545 runs in each of the three intentions (HI, LO, and BL) . The


9/19

Linear Pendu lum Exper imen tT A B L E 1

Full Database, Differences by Major ParametersHI - LO Com parisons and INT - B L Comparisons

479

Subset N-Pairs Dif f SD Z, HI-LO p-value Z-INT-BL p-valueA l l D a taLocalR em ot eFemaleMaleVolitionalInstructedFull FeedbackOther FeedbackFemale LocalMale LocalRemote FemaleRemote MaleFemale Volit.Male Volit.Female Inst rMale Inst rFull Fbk FernFull Fb k MaleOther Fbk FemOther Fbk MaleColor Fb kDig i ta l FbkNo Fbk

154591563077 077 542149466624960930616146 931310829 6198377289232171442778

1.7671.2942.456-0.7924.310-7.4708.7625.004-8.630-3.44810.7309.2560.121-9.520-1.5282.97317.4161.7659.229-11.91836.244-0.088-19.753

-20.550

97.5100.892.4106.188.0101.999.597.4109.2106.488.2104.987.7109.177.5103.293.1103.688.6110.579.581.9213.9101.0

0.7130.3880.667-0.2071.362-1.5021.9531.325-1.244-0.7992.1181.1160.030-1.540-0.2050.4952.6060.3311.764-1.6361.760-0.013-0.474

-1.773

0.2380.3490.252(0.418)0.087(0.067)0.0250.093(0.107)(0.212)0.0170.1320.488(0.062)(0.419)0.3100.0050.3700.039(0.051)0.039(0.495)(0.318)

(0.038)

1.3671.2450.6420.6501.3401.2070.5720.9930.7510.6641.3530.1360.6821.0540.592-0.1281.2380.3171.2620.6560.5220.486-0.5071.563

0.0860.1070.2600.2580.0900.1140.2840.1600.2260.2530.0880.4460.2480.1460.277(0.449)0.1080.3760.1040.2560.3010.313(0.306)0.059

incomplete series, concatenated as sets of runs , are included in the analysessince they are viable data in all other aspects of protocol. One or more serieswere com pleted by 42 operators, 21 female and 21 male. Of these, 40 opera-tors generated at least one local series, and 12 operators, including the twowho were unable to produce local databases, completed one or more seriesfrom remote locations.Full Database

The database can be separated in to several subsets taken under differentcondit ions. There are 915 local and 630 remote runs, and an approximatelyequal numbe r of runs by male an d female operators (775 an d 770, respective-ly). Com parisons can also be made between th e volitional an d randomly in -structed modes for ass ignment of in tent ion, an d between different feedbackmo dalities. Table 1 sum ma rizes the results for the complete database brokendown by location, sex, type-of-instruction, an d type of feedback. The feedbackcomparison is between "full" (color plus digital) vs "other" (digital alone,color alone, or none) . These subsets are fur ther subdivided into male and fe-male subsets for location, type-of- ins t ruc t ion, and feedback modality. T hetable shows the number of pairs (N-pairs), their mean difference (Diff), th e


10/19


Figure 4: Cumulative Deviation of Intentions (A H Data)

standard deviation of the distribution of differences (SD),and the Z-score forth e mean difference, with corresponding p-value. Z-score equivalents to theoriginal calculated t-scores are used for conv enien ce in making comparisons;only theZ-score and thep-value are given for INT - BL.The overall difference of HI - LO is positive, but non-s igni f icant . Severalsubsets exhibit significant differences, notably th e Instructed, th e M ale Local,th e Male Instructed, and the Male Full-feedback groups. T he Volitional vs In-structed difference appears to be very important in this database, yielding op-posite effects and a computed Z-score for the difference of 2.443. The Male vsFemale difference is quite large in the local data, with a difference Z of 2.063.This difference is reversed, though not as strong, in the Remote subset; howev-er, the reversal is heavily influenced by the large database from operator 144,as discussed in the R emote D ata section . There is also a strong difference be-tween full feedback, which yields a positive effect, and the other three feed-back options, all of which show n ull or n egative results (difference Z = 1.817).A graphical representation of the data in the form of cumulative deviationsfrom th e theoretical chance expectation displays the chronological develop-m en t of the statistical trends. The terminal value of such a cumulative devia-tion corresponds to the mean difference of the data distribution (multiplied byth e number of trials). The full database concatenation shown in Figure 4 re-sembles one-dimensional random walks with steps away from th e expectedmean in the positive or negative direction. To scale th e deviation, th e figureincludes a dotted curve showing th e locus of the 0.05 p-value for cum ulativedeviations, based on the standard deviation of the HI - LOdifferences.


11/19

Linear Pendulum E xperiment 481

Figure 5: A verage Intent ion vs B aseline (A l l D ata)The somew hat irregular accum ulation of HI LO deviations in the directionof intention is shown as a solid trace, which approaches significance in the

early part of the database, but retreats somewhat during th e later portion. Thedotted traces showing the other tw o comparisons display the striking asymm e-try mentioned earlier, in that the differen ces of HI vs B L tend to be consistent-ly positive an d correlated with intention, while the corresponding LO vs BLdifferences actually tend toward a positive deviation, opposite to in tent ion.This strong asymmetry of performance contributes heavily to the non-sig-nif icant overall result, despite th e u n u s u a l proportion of extreme scores. For22 operators, both the HI and the LO intent ion scores are higher than th e base-lines, sign ificantly so for five ind ividuals, and the scores compound to a posi-tive overall INT - BL difference that approaches significance. All but two ofth e subset differences are positive, i.e. th e intentional runs tend to be higher(show smaller damping rates) than th e baseline runs . Figure 5 shows thisasymmetry to be a consistent difference between the combined average in ten-tion and the baselin e data that accum ulates steadily over the full database.A lthough n o prediction w as made for such asymmetry, th e figure again in -cludes the one-tailed p = 0.05 envelope to provide a sense of scale. Thesestrong INT - BLresults indicate a differential effect of operator intention com-pared with baselines that is orthogonal to, and hence independent of, that pre-dicted in theprimary hypothesis addressing the HI - LOdifference.The q uestion arises whe ther the asymm etry might reflect a consistent trendwithin th e sessions, wher e the baseline, as the first member of each set, mighttypically be lower than th e intent ional run s for prosaic reason s such as changesin temperature or other envi ronmenta l variables. To address this concern, th e


12/19


Figure 6: Cumula t ive Deviat ion of I n t e n t i o n s (Volitional Data)

B L data within sessions were compared. No significant differences were foundin th e aggregated baselines from th e first set compared with th e second or thirdset, and the distribution of the signs of the differences was well within chanceexpectations based on the appropriate bino mial. In fact, between th e first an dsecond aggregate BL there is a small positive difference, th e opposite direc-tion from that required to explain the asymmetry in the intention al data.Local Data

Five of the 40 local operators (12.5%) show independently significant HI -LO differences in the direction of intention. However, as noted earlier, otherind iv iduals produced com parably s trong effects in the direction opposite to in-tent ion. Exam ination of the distribution of oper ators' Z -scores reveals that th elarge number of extreme scores in both tails leads to a standard deviation of1.255, a significant increase over the theoretical expectation ( p = 0.011), indi-cating that individual operator differences may be important in this database.Contributing to the variance increase, males have a significant positiveachievement (p = 0.017), while the female operators' HI - LO difference isn egative. This male-female difference obtains over both modes of instructionan d over both categories of feedback.The most prominent difference among subsets of the pendulum database isthat between volitional and randomly instructed runs . Figures 6 and 7 showthat both have steady accumu lation s, but in opposite directions. This differ-ence is more pronounced in data from male operators alone , and the Instructedsubset for the male operators exhib its by far the largest deviation in the entiredatabase, with a p-value of 0.005 for the HI - LO difference.


13/19

Linear Pendulum Exper iment 483

Figure 7: Cumulative Deviation of In ten t ions (Instructed Data)

Remote DataUnder the remote protocol,630 ru n s were generated by 12 operators. In theH I - LO comparison, on ly one of the 12achieved a s igni f icant deviat ion, buteight had a positive effect. This subset is severely un balanced by one operator(144) who comprises nearly half the total database, exhibiting a consistentthough non-s igni f icant negative yield. This contribution severely depressesthe overall remote effect size, although it remain s larger than that of the localruns. The combined results of the 11 other operators, in contrast, are positiveand s ign if ican t ( p = 0.048), and s ign if ican t ly different from those of operator144 ( p = 0.027). Figure 8 shows th e remote data in the cumula t ive deviation

format, an d displays its chronological development.Tw o long negative trends are apparent (approximately sets 175 - 325 and425 - 575), composed pr imar i ly of data from operator 144, which are shownseparately in Figure 9. If these data are not included, th e early trend continuesan d the remote database shows a significant HI - LO difference (Z = 1.667),with an effect size considerably larger than that of the local database. Thissub set of all remotes excluding operator 144 is shown in Figure 10.The remote data are important for both practical an d theoretical reasons.The potential vulnerabili ty of the pendulum to operator induced mechanicaldisturbances is totally obviated for the remote data, which thus provide a pro-tected subset immune to spurious influences tha t might conceivably affectlocal data. B eyond this , the appearance of effects with operators located up tothousands of miles from th e device has major implications for modeling th eanomalous correlations, especially if the effect size is commensurate with that


14/19


Figure 8: Cum ulative D eviat ion of Intentions (Remote D ata)in local experiments . In fact , although the difference is not s igni f icant , the re-mote effects are larger than those f ound in the local data. It is also wor th not-in g that th e pattern of results for the orthogonal INT - B L comparison resem-bles that of the local data, with a net positive t rend.

Figure 9: Cumulat ive Deviat ion of In ten t ions (Operator 144)


15/19

Linear Pendulum Experiment 485

Figure 10: Cumulative Deviation, Remotes (Excluding Operator 144)

A N O V A and Database StructureThe pendulum database has several potentially important factors that mayinteract with th e primary variable of operator intention, and a more rigorousassessment of their independent contr ibut ions requires a comprehensiveanalysis of variance. The experiment is formally a factorial design, with un -equal cell populations an d some empty cells. While all interactions cannot becalculated, an unbalanced multi-factor analysis of variance can be performed,yielding a model that provides a useful overview and background for the indi-vidual t-test assessments of particular questions. To avoid small cell popula-tions, the A N O V A was restricted to 200 series prod uced by "prolific" opera-

tors who have completed 25 or more sets of runs. Eighteen people have met orexceeded this criterion, and they have collectively produced a total of 1311 tri-polar sets of runs (85% of the full database) in the local an d remote protocols.The results in this prolific operator database are consistent with those of thefull database (Nelson an d Bradish, 1993).Table 2 displays th e A N O V A results, showing th e main effects and the inde-pendent con tr ibut ions from secondary parameters and their two-way interac-tions, using the HI - LOZ-scores for the series as the dependent variable. Foreach potential source of variance, th e table gives the sum of squares, the de-grees of freedom (D . F.), the mean square, and the F-ratio with its associated p-value. The residual variance is used as the error estimate for all factors.Combined contribution s from th e main effects indicate th e degree of overallstructure in the model, i.e.a relationship of the HI - LO difference to the ex-perimental parameters, and the remaining rows in the table show the specific


16/19

486 R . D . Nelson et al.T A B L E 2Pendulum A N O V A , Prolific Operator Database

SourceMain Effec tsR/V/IM /FF/OIn teract ionsR/V/ I /xM/FR/V/I/xF/OM/F/xF/OCovariateResidualTotal

Sum of Sqrs11.9477.6970.4683.7822.4691.1261.2370.1051.371198.427214.214

D F421152211189199

Mean Sqr2.9873.8490.4683.7820.4940.5630.6180.1051.2711.050

F-Ratio2.8453.6660.4463.6020.4700.5360.5890.1001.206

p-value0.0250.0270.5120.0590.7980.5860.5560.7550.255

contr ibut ions from these parameters an d their interaction s. T he secondary pa-rameters include a factor comparing three distinct instruction protocols, i.e.th e remote subset and the voli t ional an d r andomly instructed local subsets(R/V/I); a two level factor for male an d female operators (M/F); and a twolevel distinction of full feedback (both color an d digital) vs all other feedbackoptions (F/O). It should be n oted that the remo te subset does not include datafrom the V/I subset nor the F/O factor, an d hence th e R/V/I x F/O interaction isn ot influenced by the remote data. The INT - B L difference is entered as a co-variate in the primary calculation to test its orthogonality. A significant con-tribution would indicate that this difference is not independent, but covarieswith one or more factors in the mod el. In accord with th e expected orthogonal-ity, it is found to yield a n egl igible cont r ibut ion .Confi rming th e earlier ad hoc results, the A N O V A model returns a p-valueof 0.025 for the main effects, providing a clear indication of structure in thedatabase, driven by a significant contribution from differences among the vo-litional, instructed, and remote subsets, and by a strong difference of the com-bined color/digital feedback subset compared with all other feedback modes.The former contribution is primarily from a difference between the two in-struction modes, which yield opposite effects; a supplementary model restrict-ed to local data alone shows the V/I factor to be significant at p = 0.014. Thesetw o condi t ions effectively cancel each other in the ful l database and this large-ly explains th e small size of the overall deviation attributable to operator in -tention (seealso Table 1).The difference between Volitional and Instructed assignment of intent ion isstriking, an d broadly distributed across operators, both in the magni tude of theHI - LOdifference and in thedegree of asymmetry reflected in the INT - BLcomparison, but the present analysis does n ot suggest an adequate explanationfor the V/I difference. Certainly the signif icant asymmetry of the Volit ionalsubset contributes to this difference, but it does not by itself explain th e nega-tive sign of the Volitional data. O n e possibility is that certain consistently suc-


17/19

Linear Pendulum Experiment 487cessful or particularly unsuccessful operators tend to select just one of the twomodes, and, in fact, operators do express strong preferences and tend to useonly one of the modes. However, an examina t ion of ind iv idua l databases doesnot support this hypothesis. Of the prolific operators who did explore bothmodes, about 75% succeeded in the Instructed protocol, but only about 30%did so in the Volitional mode.The feedback comparison collapses three infrequently used modes, coloronly, digital only, and no feedback, into a single category for comparison with"full" feedback combining both color and digital modes, which operatorschose almost three times as often as the other conditions combined (see Table1). Separately, each of the three less f requent ly chosen feedback modes yieldsa negative n et effect, and in this A N O V A model, the F/O factor indicates amarginal ly s ign if ican t difference of these compared with the full feedback.T he male vs female factor is not a large contributor, nor are i ts interactionswith R/V/I or F/O.This appears to be inconsistent with the t-test comparisonswhich suggested a difference in male an d female performance. A separatemodel that collapses th e volit ional an d instructed datasets (making a local sub-set for comparison with th e remotes) shows a relatively strong, though n o n -significant interaction of the M/F factor with the new "location" factor. Thatis, the males tend to do better in the local cond ition, and females better underthe remote protocol. A nother separate model was computed replacing the M/Ffactor with an 18 level factor representing Operators. It showed that differ-ences among th e prolific operators do not generate a significant contributionto th e model, despite th e increased variance across all 40 operators' resultsdiscussed in section 6.2, suggesting that th e other experimental factors ac-count for some part of the apparent inter-operator variability.

Discussion and ConclusionsIn designing this experiment, one of the primary questions to be addressedwas whether an analog measurement of variations in the performance of a

physical system, together with the "analog" character of the operator's experi-ence, might reveal larger effects of h u m a n consciousness than those shownearlier in digital systems. That question has been answered in the negative.While the analog linear pendulum appears to be viable as a formal experimentfor exploration of interactions of human consciousness with physical systems,it does n ot yield stronger results than the digital experiments . Ind eed, the char-acter of the results in the pendulum experimen t closely replicates those in bothour R EG and R MC studies. A gain there is a persistent accumulation of verysmall statistical deviations in the direction of intention, contributed by manyof th e operators rather than by one or two particular individuals , and againthere is evidence of an asymmetry between the two intentions . This mar ginallysignificant asymmetry is independent evidence of an effect of human in ten-t ion, and it bears further examination to explore it s correlates. For example,th e effect is particularly strong in the volitional subset, as is true for the HI -


18/19

488 R . D . Nelson et al.LO difference as w ell, and it is stronger for the male operators than the female.Virtually all the effect derives f rom t rends opposite to in tention u n d e r the LOins t ruct ion , that is, there is an asym metry within the asymmet ry that favors theH I direction. It is ins t ruct ive that similar asymm etr ies have been observed inth e benchmark R EG experiment , an d with particular clarity in the RM C exper-imen t (Nelson, et al, 1991; D u n n e , Nelson, & J ahn , 1988).T he remote database also shows a larger effect size than th e local database ,as was found in both the REG and R MC experiments ( D u n n e an d J ahn , 1991).A direct comparison of effect sizes across experimen ts based on the num ber oftrials or bits processed is not feasible since the analog and binary event mea-sures are f u n d am en t a l l y different . However , the t ime spent by operators in te r -acting with the experiment can be used to normalize the terminal scores ofthese exper imen ts in a way that allow s a ten ta t ive comparison (Nelson, 1994).For the pendulum experiment, the combined local and remote effect sizeE(t) = zVhrs calculated for the prolific operator subset is 0.170. The corre-sponding prolific operator effect sizes for the diode-based REG and the RMCare 0.236 an d 0.251, respectively, both well within chance variation of thependulum effect size.This comparability of outcomes also has an implication for the integrity ofth e pendulum data. Since complete isolation from possible spur ious influencesis impractical, th e experiment depends on a physical and statistical de sign thatallows confidence in the data without making the exper iment cumbersome.The similarity of the pendulum results to those of the fully protected REG,along with the compara bility of the remote results, indicates that n o large con-tributions have arisen from spurious sources.A num ber of f indings in the pendulum analysis are especially revealing inthe i r similarities to those of other exper iments . Most impor tan t is the indica-t ion of structure imposed by operator i n t en t ion on a nomina l ly r andomprocess, despite non-signif icant overall correlation with the damping rate. T hegreatest con tribution to this outcom e is the significant difference between ran-domly and volitionally instructed datasets. The magni tude of this differenceand its gen erality across operators are unusual ly pronounced in the pendulumexperiment, an d fu ture experiments might profitably focus on this parameter.It is important to note that the random instruction provides full assurance thatthe HI and LO intentions cann ot be chosen to exploit an y temporal t rends, yetthis condition yields th e strongest, indeed independently significant, results.The instructed data show less of the asymmetry that characterizes the overalldatabase, helping lay to rest any concern tha t the asymmetry might reflectsome unknow n , non-anomalous aspect of the experimental protocol. Indeed, itcan be seen that the largest portion of the asym metry is due to the strongly in -verted LO - BLdata within the volitional subset.Of the four feedback options, by far the most successful is the combinedcolor an d digital feedback. It is also greatly preferred by the operators overother options. Color alone has a fairly substantial database, while the digital


19/19

Linear Pendulum Experiment 489an d no-feedback subsets are very small, but non e of these alone appears to pro-mote successful performance. It must be noted that assessmen ts in the smalldatasets are vulnerable to confounding, since the data represent only two orthree operators, an d there is modest evidence for inter-operator differences. I tshould also be n oted that the remote data , which show a larger effect size thanth e local data, have no on-l ine feedback at all,lending a fur ther cautionarynote for interpreting apparent differential effects of feedback condit ions .A lthough most operators regard th e pendulum as an attractive and enjoyableexperiment, its relatively modest yield in terms of effect size, its cumbersomedata processing , and its potential vulnerability to various non-anomalous influ-ences will probably limit future data collection to explorations by individualswho wish to develop large personal databases. These data will be used to deter-mi ne more precisely th e differential effects of optional parameters, differencesamong operators, and differences between local an d remote protocols.

AcknowledgmentsThe Princeton Engineering Anomal ies Research program is supported bygrants from the McDonnell Foundation, the John E. Fetzer Institute, TheOhrstrom Foundation, Inc., Helix Investments, Ltd., and Mr. Laurance S.Rockefeller. Special thanks are extended to the many volun teer operators w hogenerously gave of their time and energy to participate in the experim en ts.

ReferencesD u n n e , B . J . , Nelson, R. D., and J ahn , R . G . (1988). Operator-related anomalies in a random me-chanical cascade experiment. Journal of Scientific Exploration, 2, 155.D u n n e , B . J . and J ahn , R . G . (1991). Experiments in remote human /machine interaction. Journalof Scientific Exploration, 6, 311.Jahn , R . G . , D u n n e , B . J . , and Nelson, R . D . (1987). Engineer ing an omalies research. Journal ofScientific Exploration, 1, 21.Nelson, R. D . , D u n n e , B . J ., and J a h n , R . G . (1984). An REG Experiment with Large Data BaseCapability, III: Operator Related Anomalies (Technical Note PEAR 84003). Princeton Engi-

neering A no malies Research, Princeton University, School of En gineering/A pplied Science.Nelson, R. D. , D obyn s , Y. H., D u n n e , B . J., and J ahn , R . G . (1991). A na lys is o f Variance of RE GExperiments: Operator Intention, Secondary Parameters, Database Structure (TechnicalNote P E A R 91004). Princeton Engineer ing Anomal ies Research, Princeton Universi ty, Schoolof Engineer ing/A pplied Science.Nelson, R. D. and Bradish G. J . (1993). A Linear Pendulum Experiment: Effects of Operator In-tention on Damping Rate (Technical Note PEAR 93003). Princeton Engineering A nomaliesResearch, Princeton University, School of En gineering/A pplied Science. A n earlier versionwas presented at the 11th A n n u a l Convention of the Society for Scientific Exploration, Prince-to n NJ, J u n e , 1992.Nelson , R . D . (1994). E f f e c t Size per Hour: A Natural Uni t for Interpreting Anomalies Experi-ments (Technical Note PE AR 94003). Princeton Engineer ing Anomal ies Research, PrincetonUniversi ty, School of Engineering/A pplied Science.

R.D. Nelson et al- A Linear Pendulum Experiment: Effects of Operator Intention on Damping Rate

Documents

Transcript of R.D. Nelson et al- A Linear Pendulum Experiment: Effects of Operator Intention on Damping Rate