Hawthorne Studies: The Impact of Human Relations and Group Dynamics on Worker Productivity, Lecture notes of Decision Making

Download Hawthorne Studies: The Impact of Human Relations and Group Dynamics on Worker Productivity and more Lecture notes Decision Making in PDF only on Docsity! Seoul Journal of Business Volume 22, Number 2 (December 2016) Testing Human Relations Hypothesis of the Hawthorne Studies* JEONG-YEON LEE** Seoul National University Seoul, Korea Abstract Employing the method of time series analysis, this paper analyzes data obtained from the Hawthorne experiment from the perspective of human relations. Although previous studies adopted statistical tools to analyze the “first relay” experiments, direct inclusion of “human relations” variables was absent. The study includes “human relations” variables that suggest social facilitation and social learning process in the statistical analysis. Unlike previous studies, the direct inclusion of such variables resulted in the support for the human relations hypothesis. Keywords: Hawthorne studies, social facilitation, social learning process, human relations, time series analysis Testing Human Relations Hypothesis of the Hawthorne Studies The Hawthorne Studies, 1924-32 (See Roethlisberger and Dickson, 1939), are one of the best-known and most influential research studies in the field of social science (Hassard, 2011). The studies are often associated with Elton Mayo, a Harvard Business School professor who joined the research team at the Western Electric Company in Illinois in 1924. The results from the study formed the basis of the human relations approach, which challenged the principles of scientific management by Frederick F. Taylor (1911). The major finding of the studies includes that (1) * This study was supported by the Institute of Management Research at Seoul National University. ** Professor, Graduate School of Business, Seoul National University, E-mail: jaytalks@snu.ac.kr, Tel: 82-2-880-8252 26 Seoul Journal of Business behavior and sentiments are closely related, (2) group influences significantly influence individual behavior, (3) group standards establish individual worker output, and (4) money is less of a factor in determining output than were group standards. Despite the reputation of the Hawthorne studies, when scholars later analyzed the data with modern statistical tools, the results have not been as clear as originally claimed. Frank and Kaul (1978) were the first scholars who analyzed what we know as the “first relay” experiment. Their motive to analyze the data can be seen in the following excerpt: The massive Hawthorne experiments of some 50 years ago serve as the paradigmatic foundation of the social science of work. The insights gleaned from these experiments provide a basis for most current studies in human relations as well as for subareas, such as participation, organizational development, leadership, motivation, and even organizational design. But aside from visual inspection and anecdotal comment, the complex of data obtained during the eight years of the Hawthorne experiments has never been subjected to thorough-going scientific analysis. (p 623) The Hawthorne experiments, as they put it, became the foundation of the field of human relations by providing the following conclusion: Instead of measured experimental variables, such as physical conditions and economic incentives scheme, the unmeasured quality of human relations between workers and management and among peer groups was responsible for the overall output improvement of worker productivity. Interestingly and disputably, what Franke and Kaul (1978) found in their analysis was the opposite of what the original Hawthorne researchers described. Using stepwise regression, Franke and Kaul identified three factors that explained 94.48% of the variance when output is measured by hourly output: (1) managerial discipline, (2) economic depression, and (3) scheduled rest time. These external factors rather than internal factors such as human relations are key to the increase in productivity. These factors were left in the equation to explain worker productivity after stepwise regression. A l though Franke and Kaul made an ad justment f or autocorrelation in their analysis, their use of stepwise regression casts doubts on whether they treated the human relations Testing Human Relations Hypothesis of the Hawthorne Studies 29 METHOD Sample In total, 270 weeks were available. The original work of Franke and Kaul (1978) includes only 23 periods in which the total weeks of the experiment are somehow collapsed into groups. But, Franke (1980) lists weekly data for hourly output productivity. Because of the equal duration (week) for each time point, weekly data is used in this analysis. However, the weekly data has missing data during the 1st to 3rd week and the 251st to 270th week. Thus, the data during these periods are deleted listwise. The 67th, 68th, 117th, 118th, 169th, 170th, 221st, and 222nd weeks were also deleted since these were vacation periods. Also, there were 11 missing cells (47th ~50th and 63rd ~ 71st weeks) in terms of voluntary rest time for Worker 1, so these values were replaced with the average (=6.2) of all the voluntary time of Worker 1. In total, 239 time points of data were entered into the final analysis. Measure The hourly output (HO3). Originally, the Hawthorne studies track the hourly output of five workers over nearly five years. In my analysis, Worker 3 is the focus of the analysis for two reasons: (1) Worker 3 shows a middle level of productivity over the experimental period, and (2) Worker 3 has never been replaced by managers. The hourly output of the most productive individual (HO2). The most productive individual is Worker 2, who showed consistently better output over the 243 weeks. The group’s hourly productivity (HOAV). The group productivity level that can function as group pressure to Worker 3’s hourly output is measured by the average work output that excludes Worker 3’s output. Managerial Discipline (MD). As in Franke and Kaul (1978), managerial discipline is a dummy variable that indicates the replacement of two of the five workers (codes as 1; 0 otherwise). 30 Seoul Journal of Business Economic depression (ED). As in Franke and Kaul (1978), economic depression is measured by a categoric nature (1=economic depression; 0 otherwise) Scheduled rest time (SRT). As in Franke and Kaul (1978), scheduled rest time is measured by rest time measured by minutes, time which is scheduled by managers. Analysis The model is set up using the two input variables that represent human relations (HO2 and HOAV) and three control variables (MD, ED, SRT) that represent the external factors of Frank and Kaul (1978). Since it is time series data, the lagged input for HO2 and HOAV (that are, HO2t-1 and HOAVt-1) is also entered as well as the autoregressive term(s) of hourly output of Worker 3 (HO3t-1). Using a linear stochastic difference equation model, the parameters (represented by α and β) are associated with these terms. The model is represented in equation 1. HO3t = α + β1・HO3t-1 + β2・HO2t + β3・HO2t-1 + β4・HOAVt + β5・ HOAVt-1 + β6・MD + β7・ED + β8 ・STR (Equation 1: First order assumed, more Orders are possible) RESULTS Figure 1 shows the weekly time series of hourly output of five workers and the two workers, the most productive worker (Worker 2) and Worker 3. It shows an upward trend overall for all the workers, which allowed the Hawthorne researchers to conclude the mysterious productivity increase despite various external experimental factors. However, Franke and Kaul (1978) raised an issue regarding the lack of statistical analysis of the time series, which made the conclusion questionable. First, a descriptive analysis was done to see the nature of the given time series. Figure 1 shows the pattern of the hourly outputs of five workers (figure 1a) and Worker 2 and 3 (figure 1b). Testing Human Relations Hypothesis of the Hawthorne Studies 31 0 10 20 30 40 50 60 70 80 90 0 50 100 150 200 250 300 HO1 HO2 HO3 HO4 HO5 0 10 20 30 40 50 60 70 80 90 0 50 100 150 200 250 300 HO2 HO3 Note: HO1: Hourly output of worker 1 HO2: Hourly output of worker 2 HO3: Hourly output of worker 3 HO4: Hourly output of worker 4 HO5: Hourly output of worker 5 Figure 1. Hourly Output of Five Workers over 243 Weeks a. Five worker outputs b. Worker 2 and 3 outputs 34 Seoul Journal of Business Cross-correlation The cross-correlation output (maximum lag = 238) is presented in figure 4. The figure is based on rHO3∙HO2 (k), which indicates an error correlation of HO3 and HO2 assuming HO3 leads HO2 with lag k. The right half of the diagram indicates rHO2∙HO3 (k), since lag – k means HO2 leads HO3 instead of the other way around. As seen in the figure, approximately up to lag 70, rHO3∙HO2 (k) is less than rHO2∙HO3 (k), which indicates HO2 leads HO3. This is consistent with our hypothesis that the most productive worker (Worker 2) leads Worker 3 who is the worker of about the average of productivity in the group. -250 -200 -150 -100 -50 0 50 100 150 200 250 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 lag C ro ss co rre la tio n Crosscorrelation between Hourly Output of of Worker3 and Most Productive Individual -250 -200 -150 -100 -50 0 50 100 150 200 250 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 lag C ro ss co rre la tio n Crosscorrelation between Detrended Hourly Output of of Worker3 and Most Productive Individual Figure 4. Crosscorrelation Plot (maximum lag = 238) Testing Human Relations Hypothesis of the Hawthorne Studies 35 When the cross-correlations are further investigated in shorter lags (maximum lag =10), the plot (figure 5) shows an interesting pattern. For some reason, at lag 2, rHO3∙HO2 showed a significantly high jump pattern although the left side of the plot (which indicates rHO2∙HO3 (k)) overall is bigger than right side of the plot (rHO3∙HO2 (k)). This might indicate that at lag 2, the hourly productivity of Worker 3 (HO3) might lead the worker with the highest productivity (HO2); but overall, at other lags the other way holds true. This possibly indicates worker interdependence. -10 -8 -6 -4 -2 0 2 4 6 8 10 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 lag C ro ss co rre la tio n Crosscorrelation between Hourly Output of of Worker3 and Most Productive Individual -10 -8 -6 -4 -2 0 2 4 6 8 10 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 lag C ro ss co rre la tio n Crosscorrelation between Detrended Hourly Output of of Worker3 and Most Productive Individual Figure 5. Crosscorrelation plot (maximum lag = 10) 36 Seoul Journal of Business Spectral analysis Since the total number of observations during 239 weeks were 239, the frequency in the analysis will be 1 cycle/week. The Niquist frequency will be 120 cycles/240 weeks, which is .5. Deviations from a flat spectrum indicate some type of autocorrelation. Figure 6 shows periodograms for original and detrended data for the hourly output of Worker 3 and the most productive individual. 0 20 40 60 80 100 120 0 1000 2000 3000 4000 P ow er Frequency of Hourly Output of Worker3 Original Spectrum 0 20 40 60 80 100 120 0 1000 2000 3000 4000 5000 Frequency of Hourly Output of Most Productive Individual P ow er 0 20 40 60 80 100 120 0 0.2 0.4 0.6 0.8 1 1.2 1.4 Po we r Frequency of Detrended Hourly Output of Worker3 Original Spectrum 0 20 40 60 80 100 120 0 0.2 0.4 0.6 0.8 1 1.2 1.4 Frequency of Detrended Hourly Output of Most Productive Individual Po we r Figure 6. Spectral Analysis for Original and Detrended Data Testing Human Relations Hypothesis of the Hawthorne Studies 39 per year. In other words, the effect of a year or a half-year seems to be the reason for this systematic frequency although the power of the year is stronger than the half-year effect. A similar kind of pattern was shown for the most productive worker’s hourly output, but there was one more bump at 12 cycles/240 weeks. That is about 2.4 cycles per year. However, it is hard to identify what causes such systematic cycles for that frequency. The spectral plot filtered through the frequency domain shows a different pattern. Both for Worker 3 and Worker 2, the frequency level of approximately 48 cycles/240 weeks, which is equivalent to 9.6 cycles per a year, showed a possible significant contribution to the variance. Coherence Analysis Coherence is the covariance between the amplitudes of the two series at a frequency. Figure 9 shows that there is strong coherence in a high frequency. That means there is high covariance between the amplitudes of the two series, hourly output of Worker 3 and the most productive worker, at a high frequency. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 Frequency C oh er en ce F un ct io n E st im at e Figure 9. Coherence Plot 40 Seoul Journal of Business Model estimation I started the model testing based on Hypothesis 1 and 2 by putting contradicting theoretical variables (external factors vs. human relation factors) separately into the regression equation. This is to get an idea of how much variance each theoretical set of variables can solely explain the variance of Worker 3’s hourly output. Table 1 shows the result of the analysis. The R-square that is explained by the external factor model and the human relations model are .51 and .53, respectively. Without having control variables, this shows that the variance explained by each set of theoretical terms seems to be about equal when the effects of the human relations variable are lagged by one week. The result shows that conflicting conclusions are possible in interpreting Worker 3’s output. However, the Durbin-Watson statistics of the external factor model suggests that there is indeed error correlation (DW=1.66), indicating that further time series analysis is needed. To test the hypotheses and model testing, first order, second order and third order models that include both human relations factors and external factors that can be considered control variables were tested. The result is summarized in table 2. Overall, the R-square explained by the three models improves significantly when the order of the model increases (R2 = .55, R2 = .77, R2 = .83). The F-differential test cannot be performed since the three models are not exactly nested due to the loss of one sample size by the lag effect. The first data point of a variable when lagging the variable will be missing, and the listwise deletion reduces one sample size that includes the missing variable as the lag increases by one. However, a dramatic increase in the R-square and F-statistics suggests that the best model to explain the given hourly output of Worker 3 is third order model. In addition to the R-square and F-statistics, the estimated parameters included in model 3 overall were statistically significant compared to the other two models. Out of the three autoregressive terms, Worker 3’s hourly output at lag 1, lag 2, and lag 3, hourly output at lag 1 and lag 2 were significant (p < .001 and p <.01, respectively). What is interesting is that the output of Worker 3 at lag 2 has negative coefficients meaning that it is negatively related with the current output of Worker 3. In other words, the output of the last week is positively related with that of the current week, but Testing Human Relations Hypothesis of the Hawthorne Studies 41 Ta bl e 1: M od el t es ti ng U si ng C on tr as ti ng T he or et ic al V ar ia bl es V ar ia bl e La be l Ze ro O rd er , E xt er n al F ac to rs O n ly Fi rs t O rd er , H u m an R el at io n s V ar ia bl es O n ly Pa ra m et er E st im at e S ta n da rd E rr or t V al u e Pa ra m et er E st im at e S ta n da rd E rr or t V al u e Te rm In te rc ep t In te rc ep t 50 .9 3 0. 92 55 .3 9* ** 19 .9 3 3. 27 6. 09 A R (1 ) H O 3_ 1 W or ke r 3’ s H ou rl y O u tp u t at la g 1 0. 18 0. 06 2 .8 1 * * M A (0 ) H O 2 M os t Pr od u ct iv e W or ke r’ s C u rr en t H ou rl y O u pu t -0 .0 2 0. 08 -0 .2 1 M A (1 ) H O 2_ 1 M os t Pr od u ct iv e W or ke r’ s H ou rl y O u pu t at la g 1 0. 06 0. 08 0. 76 M A (0 ) H O A V G ro u p’ s C u rr en t H ou rl y O u tp u t 0. 47 0. 14 3 .3 5 * ** M A (1 ) H O A V _1 G ro u p’ s H ou rl y O u tp u t at la g 1 -0 .0 4 0. 14 -0 .2 6 M A (0 ) M D M an ag er ia l D is ci pl in e 7. 76 0. 89 8. 68 ** * M A (0 ) E D E co n om ic D ep re ss io n 2. 21 0. 66 3. 34 ** M A (0 ) S R T S ch ed u le d R es t Ti m e 0. 20 0. 04 4. 94 ** * F 8 0. 49 ** * 5 1. 85 ** * R -s qu ar e 0. 51 0. 53 A dj u st ed R -s qu ar e 0. 50 0. 52 D u rb in W at so n 1. 66 2. 07 Fi rs t O rd er C or re la ti on 0. 17 -0 .0 4 * p <. 05 ; * * p < .0 1; * ** p < .0 01 . 44 Seoul Journal of Business study suggests the possibility that without the inclusion of human relations variables in the statistical model, the conclusion of such statistical analyses can be misleading. Based on this possibility, social facilitation and the social learning process were identified as the primary underlying process of the human rela tions hypothesis in the Hawthorne Studies. By employing a more sophisticated time series analysis, it is concluded here that the group and the most productive individuals motivate and exert pressure on an individual’s output over time. When these variables were considered in the model, surprisingly, external factors, such as economic depression, managerial discipline, and scheduled rest time had little effect on the output level of a worker when these human relations variables were taken into account into the time series model. Thus, the model testing included in the paper supports the notion of human relations from the Hawthorne Studies (Roethlisberger and Dickson, 1939). The current study contributes to the literature in two distinct ways. First, it theorizes social facilitation (Guerin, 1993) and social learning (Bandura, 1971) as the underlying process of the human relations hypothesis. Secondly, methodologically, it employs time series analyses that capture the wave effects of these variables. The approach is much more adequate than the previous studies. The study, however, is not without limitations. Due to missing data, conclusions on all other workers were not possible. If more advances in statistical analyses are available, perhaps further investigation that analyzes other worker’s output level other than that of Worker 3 may be possible in the future. This can be helpful to validate the conclusion provided here. Despite the potential limitations, this study suggests that the previous statistical analyses on “first-relay” experiment of the Hawthorne study did not properly test “Human Relations” hypothesis. When variables representing group influences and interpersonal interaction (i.e., social facilitation and social learning) were included in the model, these variables explained the significant portion of the individual worker productivity (Worker 3’s productivity), supporting the notion of human relations. Testing Human Relations Hypothesis of the Hawthorne Studies 45 REFERENCES Allport, F. H. (1924), “Response to Social Stimulation in the Group,” in Social Psychology, F. H. Allport, ed., Erlbaum (Hillsdale, NJ), 260-291. Bandura, A. (1971), Social Learning Theory, General Learning Press (New York). Franke, R. H. (1980), “Worker Productivity at Hawthorne,” American Sociological Review, 45(6), 1006-1027. Franke, R. H., & Kaul, J. D. (1978), “The Hawthorne Experiments: First Statistical Interpretation,” American Sociological Review, 43, 623-643. Guerin, B. (1983), “Social Facilitation and Social Monitoring: A Test of Three Models,” British Journal of Social Psychology, 22, 203-214. Guerin, B. (1993), Social Facilitation, Cambridge University Press (Boston, MA). Guerin, B., & Innes, J. M. (1982), “Social Facilitation and Social Monitoring: A New Look at Zajonc‘s Mere Presence Hypothesis,” British Journal of Social Psychology, 21, 7-18. Hassard, J. S. (2012), “Rethinking the Hawthorne Studies: The Western Electric Research in Its Social, Political, and Historical Context,” Human Relations, 65 (11), 1431-1461. Jones, S. R. G. (1992), “Was There a Hawthorne Effect?” American Journal of Sociology, 98(3), 451-468. Roethlisberger, F & Dickson, W. (1939), Management and the Worker, Harvard University Press (Cambridge, MA). Taylor F. (1911), Principles of Scientific Management, Harper (New York, NY). Triplett, N. (1898), “The Dynamogenic Factors in Pacemaking and Competition,” American Journal of Psychology, 9, 507-533. Zajonc, R. B. (1965), “Social Facilitation,” Science, 149, 269-274. Zajonc, R. B. (1980), “Compresence,” in Psychology of Group Influence, P. B. Paulus ed., Erlbaum (Hillsdale, NJ), 35-60. Received September 14, 2016 Accepted October 18, 2016