Appropriate experiments were designed and carried out to verify the efficiency of the proposed CoMoRe method. This section presents the graphical and numerical results of experimental testing. The application of the method is studied, and the advantages that arise from the experimental results are analyzed. The developed simulation program uses the Python programming language in the Scientific Python development environment (Spyder) on Anaconda Navigator.
According to change models, human behavior varies. As presented in
Section 3, it starts with denial, which gradually grows and then manifests as increasing resistance. Then, acceptance slowly begins, and finally, the performance significantly increases.
The experiments involved a hypothetical company that is implementing a change. In the first experiment, the company uses 10 workers for 10 corresponding tasks (). In the second experiment, the number of workers and tasks are increased to in order to study the performance and time for larger values. The period of change, in which human behavior in response to change was simulated, was split into 15 intervals () for both experiments.
The performance values are measured as a percentage of the maximum initial performance, which is always assumed to be 100%.
Figure 3 [
51] shows the simulated curve obtained with the change management model, which reflects simulated human behavior in the phase of change. Although the simulated curve in
Figure 3 shows the typical behavior of employees in a period of change, the specific values of performance are not always constant. This uncertainty in performance values was modeled in the experiment by applying a symmetric triangular probability distribution to the values of performance and by repeating the experiment numerous times by using these random performance values. The triangular distribution was selected because for the phenomenon described by the change curve, only the average performance decline is known, having been derived from many observations in the past. It is also known that these changes are not abrupt, i.e., the standard deviation of the distribution is relatively small, and the values become less probable as we move away from the mean. Furthermore, all tailed distributions are rejected because they could lead to negative performance, which is not expected for this phenomenon. Thus, the triangular distribution is the easiest and best way to describe the uncertainty in resistance to change.
Based on the simulation of the CCMM change curve (
Figure 3), employee behavior was simulated in this experiment using the parameters in
Table 1 [
51], which shows the minimum and maximum values of resistance for every time interval, from
t = 0 to
t = 15. The interval limits are considered the lower and upper limits of the triangular distribution, which provided random values for resistance that were used during repetitions of the experiments. Red indicates that the resistance values are positive, meaning employees have a negative attitude toward the change. Green indicates “negative resistance”, which, in fact, means employees have a positive attitude toward the change. By using these resistance values and Equation (
9), the corresponding performance values can be calculated for the next interval of time.
4.1. First Experiment: Algorithm Implementation
The application of the algorithm starts at time
, along with the initiation of the change period. Following the algorithm flowchart presented in
Figure 2, in step 1, a
performance matrix
contains the initial values. These values are randomly selected from the range between 90% and 100% based on assumption (a) in the problem description in
Section 3.1 (each worker will be able to carry out all available tasks). Consequently, each worker’s performance in each possible task takes random values from 90 to 100 as an initial condition of the algorithm application. An example of such random values is shown in
Table 2.
In step 2, the Hungarian algorithm is employed to identify the task that each employee will perform to achieve the overall maximum performance. The optimal human resource allocation is demonstrated in
Table 2. The bold numbers correspond to the optimal allocation.
The third step of the algorithm is to calculate the total performance at
according to the optimal human resource allocation.
Table 3 simplifies
Table 2 by presenting only the optimal allocation of human resources to tasks, the corresponding performance values of the workers, and the total performance, which is calculated as the sum of all performance values. In a hypothetical scenario where all ten workers exhibit maximum performance levels in all tasks, the maximum total performance would be 1000 (
). However, in this example, the optimal allocation results in a total performance of 980.
At this point, the resistance to change is not yet apparent, and the calculated total performance is the initial state of the algorithm. This optimal allocation and the corresponding total performance will be used as the starting point for both the proposed method and the CCMM to facilitate a fair comparison between the two methods.
In step 4, the resistance to change emerges, leading to a reduction in employee performance. A
resistance matrix (
) is created for each employee and task, with random values in the range from 0 to 20, according to
Table 1, for
. An illustrative example of these random values is provided in
Table 4.
Subsequently, in step 5, a new performance matrix for the next time interval (
t = 1) is calculated using Equation (
9). Steps 2 to 6 will be repeated until the end of the change period (
t = 15). In all these iterations, the Hungarian algorithm is used to reallocate human resources to tasks, and the total performance is calculated again. New resistance matrices are created with random values in a range according to
Table 3, from
to
. A new performance matrix is calculated in every iteration according to Equation (
9).
In step 7, when the change period ends (t = 15), the Hungarian algorithm is used one last time to reallocate human resources to tasks. Finally, in step 8, the total performance is calculated according to the last human resource reallocation.
The main innovation of the proposed method (CoMoRe) compared to the CCMM is that it monitors and provides decisions for reallocating human resources to tasks during the change period. On the contrary, the classical method keeps workers in the same initial positions throughout the change period, so it cannot take advantage of new information about people’s ability to accept change during the change period.
An example of reallocations during the iterative process (from
t = 0 to
t = 14) is shown in
Table 5. For example, worker 2 (
) initially works at task 2 at
and
and then works at task 1 at
and at task 4 from
until the end of the change period. The employees that perform a specific task can also be observed. For example, task 2 (
) is initially performed by worker 2 at
and
, then by worker 9 at
and
, and finally, by worker 6 from
until the end of the change period. Reallocations are more frequent in the early stages. Bold numbers in
Table 5 indicate when human resources remain at the same task until the end of the change period.
4.2. First Experiment’s Results and Analysis
In order to test the proposed CoMoRe method, we ran the algorithm repeatedly, trying to minimize the possibility that the results were influenced by random values. These random values were used for the initial state of the performance and resistance matrices used during the iterations. Starting with 100 repetitions, the algorithm was also repeated 1.000, 10.000, and 100.000 times. The average performance was calculated for each case. The results of 10.000 repetitions were not significantly different from those of 100.000 repetitions, so there was no need to further increase the number of repetitions.
Figure 4 compares the average results of the CoMoRe method with the results of the CCMM based on the classical change curve.
Table 6 summarizes the corresponding arithmetic values used to create the curves in
Figure 4.
Figure 4 and
Table 6 show that the CoMoRe method achieves a higher total performance than the CCMM method, and the advantages are obvious. The graphical representation and corresponding arithmetic values indicate that the total performance with the CoMoRe method is better at all time intervals of change. Moreover, the drop in total performance is much lower. As shown in
Figure 4a–d, both methods start from the same value, which is the average total performance for 100, 1.000, 10.000, and 100.000 repetitions, respectively. As observed in these figures, the improvement occurs mainly during early time intervals and is maintained until the end of the change period. Another useful observation from the curves is that when using the CoMoRe method, the total performance is always significantly better than that obtained with the classical method; this improvement in performance favors a commitment to change and reduces the likelihood that the change process will be abandoned, which is often observed in companies and organizations when performance drops to significantly low levels.
Table 6 shows that in all experiments (100, 1.000, 10.000, and 100.000 repetitions), the average total performance starts (
) at a value of 981 in both methods. From
t = 1 to
t = 7, the overall performance for the proposed CoMoRe method begins to be significantly better than that for the CCMM method. At
t = 7, the lowest total performance has a value of 491 for CoMoRe versus 286 for the CCMM. This is the most critical point in the change process. At the next time step, the performance starts to increase. At
t = 14 for 100.000 iterations, the total performance for the proposed CoMoRe method is 1091 and has exceeded the initial total performance (981), while the total performance for the CCMM is less than the initial total performance, with a value of 886. Thus, the new desired state that improves performance over the initial state is reached in a shorter time.
Table 7 compares the Total Performance Improvement (TPI) in percentage between the two methods at all time intervals. The Total Performance Improvement is defined as follows:
where
and
denote the total performance with the CoMoRe and CCMM methods, respectively.
As presented in
Table 7, the lower total performance at time
t = 7 and
t = 8 is approximately 71% higher for the CoMoRe method compared to the CCMM method. As a result, the plans for change are more likely not to be rejected in the early stages, something that usually happens in the change management process. From the above data, it is evident that a specialist change process manager has the tools to deal with the RtC effect as a consequence of change and make optimal decisions. The CoMoRe method provides the manager with information that can serve as an early warning about the process of change over time, enabling the timely reallocation of human resources to achieve a successful change. Generally, a continuous evaluation of change is achieved using the CoMoRe method.
Figure 5 shows the average number of employees that will need to be reallocated during the change process (these results were produced in the experiment with 100.000 repetitions). At the early stages of change, when the RtC is high, the reallocation frequency is higher to avoid a significant drop in performance. Afterward, fewer adjustments have to be made, and the number of reallocations decreases over time. Another interesting result in
Figure 5 is that human resources are allocated to their most productive positions at about the halfway point (
t = 8) of the total change period, and no further reallocation is required.
4.3. Second Experiment: Algorithm Implementation
In the second experiment, the hypothetical company uses 100 workers for 100 corresponding tasks (
) in order to evaluate the effectiveness, efficiency, and computational complexity of the proposed CoMoRe method on a more extensive dataset. The change period, in which human behavior in the change phase was simulated, remained at 15 intervals (
). Following the same process as in the first experiment, the algorithm was applied for 100, 1.000, 10.000, and 100.000 repetitions. The results of the total performance obtained using the CoMoRe and CCMM methods are compared in
Table 8.
Table 8 shows that in all experiments (for 100, 1.000, 10.000, and 100.000 repetitions), the average total performance starts (
t = 0) with an initial value of 9900 in both methods. From
to
, the overall performance obtained with the proposed CoMoRe method begins to be significantly better than that achieved with the CCMM method. At
(for 100.000 repetitions), the lowest total performance has a value of 6560 for CoMoRe versus 2950 for the CCMM, which is a critical value for the change process. In the next time step, the performance starts to increase. At
(for 100.000 repetitions), the total performance for the proposed CoMoRe method is 10070 and has exceeded the initial total performance (9900), while the total performance for the CCMM is 5850, significantly less than the initial total performance. Thus, the new desired state, which improves performance over the initial state, has been reached in a shorter time (
), representing a 20% time reduction.
The results of the Total Performance Improvement (defined in Equation (
10)) with the CoMoRe and CCMM methods are compared in
Table 9 below.
As demonstrated in
Table 9, the CoMoRe method provides a significant improvement compared to the CCMM method in all cases. At the critical time periods of
and
, where a lower total performance is observed, the CoMoRe method achieves a significantly better result, approximately 122% improved compared to the CCMM method.
It is also important to emphasize that in the second experiment with the larger allocation problem (
), the improvement rates were higher than in the first experiment (
). This can be attributed to the larger set of workers and tasks, which provided more opportunities for optimal allocation, consequently leading to a substantial enhancement in overall performance, as evidenced in
Table 8 and
Table 9.
The CoMoRe algorithm is a computational approach that involves solving the HRAP through the utilization of the Hungarian algorithm. According to the literature, the Hungarian algorithm is known to exhibit a complexity of
. In its remaining parts, the CoMoRe algorithm does not present a computational complexity of a higher order. More specifically,
Table 10 presents the execution time of the Hungarian part of the algorithm, the time that corresponds to the remaining part, and the total execution time of the CoMoRe algorithm. It is evident from the data presented in
Table 10 that the Hungarian part of the algorithm for a
matrix needs only 28.25 s, which is 3.6% of the total execution time, but for a
matrix, it needs 826.98 s, which is 46% of the total execution time. Therefore, the execution time of the Hungarian part increases by over 2800% as the matrix size increases from
to
. The remaining part has a small increase of approximately 28.7%. The total time shows a modest increase of approximately 130%. In conclusion, while the Hungarian algorithm constitutes the primary computational issue, the overall complexity of the method remains manageable for moderate matrix sizes. The reported runtimes were measured when running the algorithm on a computer with the following characteristics: Intel® Core™ i7-8700 CPU @ 3.20 GHz, RAM 32.0 GB.