5.1 Lean versus operations financial performance (L-O)
5.1.2.1 Lean archetype versus operations financial performance
Two statistical analyses were performed in order to test the supposition that leanness--as captured in archetypal classification--is related to operations performance. The first analysis examined the relationship between lean classification and eight individual measures of operations financial performance. Table 4.13 in the Methods Chapter presents the analysis results. Figures 5.2 and 5.3 plot operations financial performance by cluster membership. The median z-scores may be interpreted as measures of distance above or below the competitive median performance for that particular metric. Wilcoxon rank-sum tests of median z-scores for lean and non-lean firms are not significantly
different for asset or employee productivity. The difference was in the expected direction for asset productivity, with the median z-score for lean being higher (0.37 versus 0.31).
Employee productivity lay in the unexpected direction with the median z-scores for non- lean firms being higher. The amount of emphasis to place on these directional
observations is limited since statistical tests did not find a significant difference. The effect of sample size on the likelihood of not finding a significant difference in
performance between lean and non-lean companies when one truly exists (Type II Error) was analyzed by resampling the original with replacement to obtain sample sizes of 100,
200, and 300. The results are reported in Table 5.1. The difference in asset productivity is only detectable at a z-score difference 3 times that found in the original sample (0.18 vs 0.06) and sample size of 100. Employee productivity is not found to be significantly different at any difference or sample size using this method.
Gross margin ratio was found to be moderately significant but also in the unexpected direction. Lean companies tend to have narrower gross margins than non-lean companies do. As discussed in the Methods Chapter, this outcome might have been foreseen. A potential reason for selecting lean as a strategic improvement program is profit margin pressure.
Where lean companies significantly differentiate themselves is with respect to cycle time performance. They have significantly lower aggregate cycle time as measured by cash- to-cash cycle time and component-level cycle time as measured by inventory and receivables performance. Total cycle time, as another aggregate measure, is also
moderately significantly better for lean companies. Payables cycle time is not significant although the difference is in the unexpected direction in that non-lean companies pay their vendors faster than lean companies.
The foregoing analysis of multiple two-sample tests must be adjusted for the increased likelihood of Type I Error. Type I Error is the likelihood that a difference will be found when one does not truly exist. The adjusted familywise error rate for eight tests
performed at the α = 0.05 level is calculated as (Keppel, 1991):
337 . ) 05 . 1 ( 1 ) 1 (
1− − = − − 8 =
= c
FW α
α
However, the chances of making a Type I Error on three of eight tests at the α =.05 level plummets to .006. It is highly unlikely that three or more of the eight tests would find a difference if one did not exist31. In this case, significant results were found for CTC, CTI, and CTR at the 0.5 level as well as differences at the 0.1 level for GMR and CTT.
Even if allowances were made for the fact that CTI and CTR are components of CTC and CTT, this result would be highly unlikely if a difference truly did not exist.
The second analysis, to avoid the problem of underestimating the probability of Type I Error, examines multiple measures of operations financial performance simultaneously.
Logistic regression was used to determine whether or not a relatively diverse and
parsimonious set of variables predicts lean or non-lean categories. Again, the results are reported in the previous methods section (ref. Figure 4.5a and b). Controlling for shared variance in this way indicates that the operations financial performance variable set is moderately significant in predicting lean classification when asset productivity, gross margin, and either cash-to-cash or total cycle time is included in the model. In either case, the cycle time term was the only significant variable. This logistic regression analysis, in conjunction with the preceding results of individual Wilcoxon rank-sum tests, lends support to a positive relationship between lean classification and operations
financial performance in the form of lower cycle times.
31 Calculation base on the binomial formula for P(X>/=3) at n=8 and α=.05.
Data
4.8 3.6 2.4 1.2 0.0 -1.2 -2.4 Cluster
ROCA
EP
GMR
L N
L N
L N
Cluster L N
Dotplot of ROCA, EP, GMR vs Cluster
Figure 5.2: Plots of lean (L) and non-lean (N) clusters’ operations financial performance (i.e. ROCA, EP, GMR).
Data
3.2 2.4
1.6 0.8
0.0 -0.8 -1.6
Cluster
CTC
CTT
CTP
CTI
CTR L N L N L N L N L N
Cluster L N
Dotplot of CTC, CTT, CTP, CTI, CTR vs Cluster
Figure 5.3: Plots of lean (L) and non-lean (N) clusters’ cycle time operations financial performance.
Difference P-value Difference P-value Difference P-value Difference P-value
ROE 0.22 0.790 0.24 0.202 0.22 0.181 0.26 0.007
SG 0.11 0.453 0.13 0.050 0.14 0.000 0.03 0.106
SR 0.24 0.905 0.23 0.484 0.30 0.563 -0.07 0.645
ROCA 0.06 0.809 0.18 0.042 0.01 0.234 0.18 0.016
EP -0.14 0.929 -0.10 0.894 -0.15 0.968 -0.10 0.110
GMR 0.30 0.077 0.16 0.115 0.24 0.001 0.35 0.000
CTC -0.65 0.040 -0.60 0.000 -0.59 0.000 -0.59 0.000
CTT -0.65 0.066 -0.44 0.014 -0.42 0.001 -0.64 0.000
Notes: 1. Nominal sample size. Adjusting for missing data: SR = 23/15, CTC = 24/17, CTT = 24/17.
2. Values based on single, simple random selection from original n=24/18 sample values with replacement to sample sizes of n=100, 200, and 300.
3. Difference based on median Z-scores lean minus non-lean.
4. P-values based on 2-tail Wilcoxon rank sum test. Values </=0.10 are in bold italics.
Business Financial Performance
Operations Financial Performance Sample Firms
(Lean/Non-lean) Single Resample with Replacement (Lean versus Non-lean)
n=24/18 (1) n=100 n=200 n=300
Table 5.1: Lean versus non-lean firm median z-score differences for the original sample and resampled data with replacement.