The effect of measurement alternatives on a nonfinancial quality measure's...
I. INTRODUCTION
Prior accounting research documents that quality-related nonfinancial performance measures such as customer satisfaction (Ittner and Larcker 1998a), defect rates, and on-time deliveries (Nagar and Rajan 2001) are often leading indicators of financial performance. Yet we
Forward-looking quality measures are important to managers because they link current quality-related decisions to future financial consequences. Conceptually, measure construction choices that strengthen the relation between a current quality measure and future financial performance increase the congruence of the measure to management objectives. The quality measure thus becomes potentially more useful for contracting (Datar et al. 2001; Feltham and Xie 1994). Moreover, current quality-related decisions are more likely to be value-maximizing when managers have a timely, yet highly congruent, quality measure by which to evaluate those decisions. The identification of measurement alternatives that can improve the forward-looking properties of a quality measure is, therefore, important to management accountants who play an important role within firms as sources of decision-useful information and as measurement experts (Kinney 2001).
Using a proprietary database of six units of a leading medical services firm, this study proposes and empirically examines whether three types of measurement alternatives affect the strength of the relation between a quality measure and future quality (i.e., warranty) costs. Drawing on theories from operations, management, and marketing to suggest measurement alternatives, the study examines: (1) the standards against which quality is measured, (2) the symmetry of the relation between the quality measure and future quality costs (i.e., whether variations in quality have a uniform affect on future quality costs), and (3) the functional form of the relation between the quality measure and future quality costs. The first speaks to the quality measure itself; the latter two speak to the form of the quality cost function.
There are three primary empirical results. First, the standard against which quality is measured affects the strength of the association between the quality measure and future quality costs. Marketing theory suggests that the customer's ex ante expectation of product performance or service outcome is the appropriate standard against which quality should be measured (e.g., Cronin and Taylor 1992, 1994; Parasuraman et al. 1985; Patterson 1993; Stank et al. 1999). However, in the current setting, an absolute quality measure (i.e., quality measured against an optimal outcome standard) is a better predictor of warranty costs than an expectations-adjusted quality measure (i.e., quality measured against an expectation standard).
Second, an asymmetry is documented in the quality (i.e., warranty) cost function in the current setting. This result is consistent with the generalizable hypothesis that the strength of the relation between the quality measure and future warranty costs is a function of customers' expected net benefits of warranty claims.
Finally, empirical results evaluating alternative functional forms of the warranty cost function do not support the tolerance limit approach that is commonly used in manufacturing settings and that assumes small variations from optimal quality are not costly for the firm. Instead, the results suggest that all deviations from optimal quality are costly in the current setting, consistent with quality theory in operations research (e.g., Taguchi et al. 1989). Moreover, the data support a linear (as opposed to an exponential) relation between the quality measure and future quality costs in the current setting.
The implication for management accounting of this study is that a quality measure can be made more useful for decision making and control by constructing the measure (i.e., with consideration of measurement alternatives such as those described above) in a way that improves its forward-looking properties. To highlight this, the paper concludes with an illustration of how firms can use an "improved" contemporaneous and forward-looking quality measure to assess quality-related decisions in a timely manner and shows that those decisions can differ when alternative forms of the quality measure are used.
Section II provides theoretical development for the quality measure and quality cost function measurement alternatives. Sections III and IV describe the research setting and the results of the empirical analysis, respectively. The decision-facilitating role of a quality measure is illustrated in Section V, and Section VI concludes.
II. THEORETICAL DEVELOPMENT
Prior research finds that quality measures (including customer satisfaction) are leading indicators of financial performance (e.g., Banker et al. 2000; Ittner and Larcker 1998a; Nagar 1998; Nagar and Rajan 2001). In this section, I draw from theories in operations, management, and marketing to examine the issues associated with measuring quality and with identifying the relation between a contemporaneous quality measure and future quality-related warranty costs (i.e., a quality cost function). First, I consider whether the choice of performance measure standard affects the strength of the relation between a quality measure and future quality costs. Second, I develop a hypothesis regarding the symmetry of the quality cost function (i.e., whether all variations in quality have a uniform affect on future quality costs). Finally, I consider alternative functional forms of the quality cost function.
Quality Standard
Quality standards are often absolutely determined. For example, precise product specifications are identified in manufacturing settings. Similarly, in a customer satisfaction response scale, the highest satisfaction level is the implicit absolute standard. Alternatively, an expectation of the quality outcome can be used as a standard against which a quality measure is constructed. For example, a firm may measure quality as the defect rate in a given period as compared to the historical average, possibly with exogenous, time-varying factors (e.g., volume) incorporated to further adjust for current period expectations.
Marketing theory suggests that a customer's overall satisfaction with a product or service (i.e., a measure of perceived quality) is a function of his ex ante expectation. That is, the gap between expectation and actual outcome determine the perceived quality (Cronin and Taylor 1992, 1994; Parasuraman et al. 1985; Patterson 1993; Stank et al. 1999). Marketing "gap analysis" documents a positive effect of perceived quality on firm profitability via repurchase (Oliver 1980; Cronin and Taylor 1992; Anderson and Sullivan 1993; Stank et al. 1999) and complaint behavior (Bearden and Teel 1983). In addition, there is some evidence that the relation between perceived quality of services and overall firm performance is positive (e.g., Phillips et al. 1983; Nelson et al. 1992; Anderson et al. 1994; Zeithaml et al. 1996).
Thus, marketing theory suggests that perceived quality measured relative to an expectation has forward-looking properties that may dominate an absolute quality measure. Since quality-related warranty work is precipitated by customers (note that this is ultimately true, even for product recalls), it follows that warranty costs are also likely to be a function of customer ex ante quality expectations. I, therefore, make the following hypothesis (stated in alternative form):
H1: An expectation-adjusted quality measure (i.e., quality measured against the ex ante expected outcome) is more strongly associated with future quality costs (e.g., warranty costs) than an absolute quality measure (i.e., quality measured against the optimal outcome).
There are (at least) two potential reasons why an expectation-adjusted quality measure may not dominate an absolute quality measure. First, if customers fixate on optimal quality as a result of, for example, promotional promises (e.g., "if not completely satisfied" type statements) then, even though ex ante expectations may be for less than optimal quality, quality measured against the optimal outcome could be more highly associated with actual future warranty claims. Second, customers may have ex ante quality expectations that are too low; for example, a firm could purposefully set customer expectations low to increase ex post perceived quality. If this becomes evident to the customer ex post, then warranty costs may be more highly associated with an absolute quality measure once the customer updates his priors.
Asymmetry
Although prior research documents that quality-related performance measures are often associated with future financial performance (e.g., Banker et al. 2000; Ittner and Larcker 1998a; Nagar 1998; Nagar and Rajan 2001), theories predicting determinants of the strength of these relations have not been proposed in the accounting literature. This paper suggests one such theory, namely, that the strength of the relation between a quality measure and future quality-related warranty costs is increasing in the expected net benefits to the customer of the warranty work. That is, the expected net benefit moderates the quality-future cost relation. Although this hypothesized moderating effect is equally plausible for manufacturing firms, it is likely to be stronger in high-contact service firms for reasons described below.
An important characteristic of service firms that distinguishes them from traditional manufacturing firms is their "joint production." Many services (especially medical service firms) require a large degree of customer contact; that is, the service is jointly produced by both the customer and the firm's employee (Kotler and Bloom 1984; Parasuraman et al. 1985; Apte et al. 1997). Because of this characteristic, it is not ex ante clear that a contemporaneous quality measure will be a leading indicator of quality-related warranty services and costs. Joint production implies higher levels of customer participation in all service deliveries. As such, the re-delivery of an initial poor quality service imposes costs on service customers, including out-of-pocket costs and opportunity costs associated with the time it takes to participate in a second service delivery. Of course, the customer also has an ex ante expectation of the benefit of the warranty work. I hypothesize that the service customer's expected net benefit of warranty work is an important determinant of whether he or she pursues resolution of the poor initial service in the form of service re-delivery.
Expected net benefits may vary depending on characteristics of the customer. For example, a customer who expects to have future transactions with the firm may assess a higher expected net benefit to the resolution of a poor-quality service than a customer who does not. Thus, one might expect the relation between service quality and warranty costs to be stronger for repeat customers than for one-time customers.
More generally, situations in which a customer's cost of re-delivery of a poor service is relatively high (or benefits relatively low) result in a lower probability of the customer pursuing warranty work. This leads to the second measurement hypothesis to be tested, which is (stated in alternative form):
H2: The magnitude of the (negative) relation between quality and future warranty costs is increasing in the expected net benefits to the customer of the warranty work.
Stated another way, future warranty costs will be more sensitive to actual quality outcomes when the customer's expected net benefits of the warranty work are relatively higher.
Functional Form
In any given setting, the relation between a measure of quality and future quality costs (i.e., the cost function) may take several functional forms, including (but not limited to) a dichotomous form as suggested by the traditional tolerance limit approach, a simple linear form, or even a quadratic form.
The traditional tolerance limit approach is widely used in product manufacturing practice and assumes that there is a range of quality outcomes that are equally acceptable. That is, as long as the product measurement outcome (i.e., a quality measure in manufacturing settings) is within predetermined "tolerance limits," regardless of how close it is to the target value, the product is deemed to be in conformance with quality standards (see Figure 1, Panel A). All products with dimensions outside of the upper and lower tolerance limits are nonconforming. The tolerance limit approach assumes that only nonconforming products result in nonconformance costs such as scrap, rework, and warranty costs (Albright and Roth 1992).
[FIGURE 1 OMITTED]
Quality management research in manufacturing settings indicates that the traditional tolerance limit approach of identifying good products is deficient because it ignores quality costs associated with small deviations from targets. Taguchi et al. (1989) argue that nonconformance quality costs will be incurred with any deviation of a product characteristic from its target (i.e., not just for products with characteristics outside of tolerance limits) (Figure 1, Panel B). (1) If this is true, then the cost function will assume a linear, or possibly quadratic, functional form.
Although Kim and Liao (1994) examine various functional forms of quality cost functions from a theoretical perspective, there is little work that attempts to discriminate empirically among these alternative forms. Identification of the correct functional form of a quality cost function provides a third measurement improvement that can potentially facilitate cost prediction and, ultimately, decision making and control. Since functional form is likely to be context-specific, I make no general predictions about the dominance of any particular functional form. Instead, in the empirical analysis I estimate and compare three functional forms: (1) a dichotomous cost function in which relatively small deviations from the quality standard are assumed to result in no increase in quality (i.e., warranty) costs, (2) a linear cost function in which all deviations are assumed to be linearly associated with increased costs, and (3) a quadratic cost function in which deviations are assumed to be associated with exponential increases in costs.
In summary, theories from the fields of operations, management, and marketing provide insight and suggest predictions regarding three quality measurement issues: (1) the alternative standards against which quality is measured, (2) whether variations in quality have a uniform affect on future quality costs (i.e., consideration of an asymmetric quality cost function), and (3) the exact functional form of the relation between the quality measure and future quality costs. The following two sections of the paper empirically examine the measurement issues described above.
III. RESEARCH SETTING AND DATA DESCRIPTION
Selection of the Research Site and Description of the Service
The research setting chosen for this study is a leading medical services firm that specializes in the surgical correction of myopic eyesight using excimer laser technology (hereafter, laser vision correction). This setting was chosen for several reasons. First, in contrast to the large amount of research on quality costs in manufacturing, there is little research of this type in service settings. The service setting chosen for this paper expands the quality research in a direction likely to be fruitful in the future (i.e., technology-driven service settings). Second, since the firm currently pursues a strategy of being a high-quality provider of laser vision correction, quality measurement issues studied are of particular importance.
Third, quality-related warranty (i.e., rework) costs associated with laser vision correction occur far in the future--over eight months in the future for the patients in the sample needing rework. Thus, a measure of quality contemporaneous to the surgical procedure would be useful to managers if shown to be a leading indicator of future costs. Finally, because of the technological nature of laser vision correction, the data are available to examine the measurement issues put forth in this paper. For example, the surgeons routinely identify an expected outcome distinct from the physiologically determined absolute quality standard (e.g., corneal curvature giving "20/20" eyesight). Indeed, a criticism of the marketing research previously discussed is the difficulty of measuring ex ante customer expectations, especially in service settings (Cronin and Taylor 1992; Parasuraman and Zeithaml 1994).
The sample consists of five surgery centers within one large firm. The centers are located in the U.S. and Canada and provide the surgical correction of varying levels of myopia (near-sightedness). The procedure involves surgically altering the shape of the patient's cornea using excimer lasers and the Photorefractive Keratectomy (PRK) procedure. The surgeries are performed at the center on an outpatient basis by an ophthalmologist who is a firm employee. The fee charged by the firm includes payment for an initial screening, the surgery, and all post-procedure exams performed by the patient's referring optometrist. The firm offers a lifetime guarantee and will provide, free of charge, an "enhancement" to patients so requiring. Thus, enhancements represent an important quality-related cost for the firm (i.e., warranty cost).
Description of the Data
For each of the centers, there are two primary data sets (Table 1). The procedure-level outcomes database contains descriptive information for 4,470 initial (i.e., not enhancement) surgical outcomes. Table 1, Panel B presents the numbers of procedures performed and the number of months of data for each of the centers. Four of the five centers began operations during the period of study, May 1996 through September 1997. (2) Center A is the most established center, having the longest time-series of data (18 months) and the largest number of (initial) procedures (3,263) performed. The monthly level database contains financial information (investments in training and maintenance), center volume, and surgeon experience for 76 center-months.
Prior to the surgery, the surgeon takes two measurements of the patient's cornea--(1) a measure of the refractive spherical curvature of the cornea, the spherical measure; and (2) a measure of the shape of the cornea, the cylinder measure--each of which is measured in "diopters." The magnitude of the spherical measure (variable MAG SPH) indicates the degree of myopia, while the magnitude of the cylinder measure (MAG CYL) indicates the degree of astigmatism. Most procedures (80 percent in the sample) include both a spherical and a cylinder correction. The two measures are combined into one measure called the "spherical equivalent." This measure, EQU, is calculated as follows:
EQU = SPH + (CYL/2). (1)
A spherical equivalent measure of 0.00 is called "PLANO." A PLANO measurement indicates no surgical correction is needed (i.e., light rays focus perfectly on the retina and there is no astigmatism) and is the optimal outcome, that is, an absolute, optimal quality standard. Note that the EQU measure is negative for myopic patients, with smaller (more negative) values indicating more severe corrections (see Figure 2).
[FIGURE 2 OMITTED]
The pre-surgery measurements, along with other factors such as patient age, are used to identify suitable candidates for the procedure and to determine the "intended" or expected spherical equivalent outcome (variable E[outcome]) used to test H1. Table 1, Panel A presents descriptive statistics for the variables used in the following two sections; Panel B presents the same statistics by center. Table 2 provides correlations (Spearman and Pearson) among the procedure level variables (Panel A) and the center-month variables (Panel B). (3)
IV. QUALITY AND QUALITY COST FUNCTION MEASUREMENT
In this section, following a brief discussion of the warranty cost proxy used in the tests, I empirically examine the three types of quality measurement alternatives identified in Section II.
Warranty Cost Proxy
A significant source of quality-related costs in the laser vision correction industry is the cost of performing "enhancements" (i.e., rework) for those patients who require them. I argue that enhancement rates capture three dimensions of quality-related costs. First, there are out-of-pocket (variable) costs of performing these procedures including the cost of medications, technical support staff, and laser fees. Second, reputation effects resulting from dissatisfied customers who required an enhancement may impose costs on the firm in the form of lost referrals. Finally, enhancements impose opportunity costs because of the inability to perform paid procedures while the enhancement procedure is being performed--a cost that is significant for centers operating at or near capacity (e.g., Center A).
The current analysis is limited by the lack of data on the explicit costs of enhancements. However, out-of-pocket costs vary little across enhancement procedures. It is, therefore, unnecessary to attempt to impute an out-of-pocket dollar cost to each enhancement. Rather, warranty cost is effectively captured by a dichotomous variable of the incidence of an enhancement. Moreover, interviews conducted with surgeons and staff optometrists reveal that more severe quality "failures" result in earlier enhancements, and that the probability that an enhancement will be given is decreasing in the amount of time that has elapsed (i.e., without an enhancement). (4) For these reasons, the incidence and timing of enhancements is a reasonable proxy for quality-related warranty costs in the current setting.
Based on the arguments given above, I use the duration of the initial procedure, calculated as the number of days from the initial procedure to the enhancement (or the end of the test period if no enhancement is performed), as the warranty cost proxy. (5) Note that DURATION is inversely related to costs; that is, high values of DURATION indicate a low likelihood of enhancement. For those patients requiring enhancements, the minimum, mean, and maximum DURATION are 118, 255, and 491, days, respectively (see Table 1, Panel A).
Comparison of Absolute and Expectation-Adjusted Quality Standards
Marketing theory suggests that the ex ante expected outcome is a more appropriate standard for evaluating quality performance than is an absolute, optimal standard. Hypothesis 1 accordingly predicts that an expectation-adjusted quality measure is more strongly associated with future quality costs (i.e., warranty costs) than an absolute quality measure. In the current setting, PLANO and E[outcome] represent the absolute (i.e., optimal) and expectation standards, respectively. Thus, the difference between the initial post-surgery spherical equivalent measure, POST EQU, and PLANO is an absolute quality measure. Similarly, the difference between POST EQU and E[outcome] is an expectation-adjusted quality measure.
I test H1 by comparing the strength of the associations between each of these measures and the warranty cost proxy, DURATION. A maximum likelihood estimation of procedure duration, DURATION, as a function of each measure is estimated. This estimation assumes a Weibull distribution for duration and adjusts for right censoring of procedures not requiting an enhancement by the end of the test period. (6)
The results of the test of H1 are presented in Table 3. Both the absolute and the expectations-adjusted measures (Models 1 and 2, respectively) are significantly (p-value < .001) negatively associated with DURATION, indicating that larger deviations from the respective standard (i.e., lower quality) result in a shorter procedure duration--that is, a higher likelihood of a future enhancement (i.e., higher future warranty costs). However, the model with the absolute quality measure has a pseudo-[R.sup.2] of 6.63 percent, while the model with the expectation-adjusted measure has a pseudo-[R.sup.2] of only 2.31 percent. Moreover, when both measures are included in the model, only the absolute measure remains significant; that is, the absolute measure is incrementally informative over the expectation-adjusted measure, but not vice versa. This suggests that, contrary to HI and prior theory, quality measured against the absolute standard is more highly associated with future warranty cost in the current setting than is quality measured against the expectation standard.
One potential explanation for this somewhat surprising finding is that customers, regardless of expectations the surgeons attempt to set, fixate on the optimal outcome of 20/ 20 vision and implicitly use PLANO as the standard. In other words, the surgeon's expected outcome, although communicated to the patient, was not the patient's expected outcome. Consistent with the previously described criticism of extant marketing gap analysis research, this highlights the difficulty and importance of measuring ex ante expectation (an additional measurement issue, but one that is beyond the scope of this paper), a common concern in the marketing gap analysis research described above.
A second potential explanation for the unexpected result is the surgeon's incentive to set an overly conservative expectation standard to increase the likelihood of exceeding this standard. Although medical professionals are, by law, restricted from being compensated based on medical outcomes, the surgeons likely have implicit incentives (e.g., reputation) to provide high quality. Because the expected outcome is determined subjectively by the surgeon, he has both the incentive and the opportunity to understate the expected quality. The above-described findings are consistent with an overly conservative estimate of the expected quality outcome on the part of the surgeon, and with the warranty claim (i.e., enhancement) decision ultimately being made based on the optimal standard of PLANO.
Asymmetry in the Quality Cost Function
The results presented above document that procedure DURATION is decreasing in (an absolute measure of) quality. In this section, I examine whether the magnitude of this relation is uniform, in this case, across customers. Hypothesis 2 predicts that the magnitude of the negative relation between quality and warranty costs is increasing in the expected net benefits to the customer of the warranty work. The current setting provides a natural means of testing this hypothesis.
Recall that the original procedures in this sample were for the correction of myopia in which the cornea is too curved and must be flattened in the center by the laser. Surgical outcomes to the right of PLANO (see Figure 2) represent overcorrections by the surgeon; the patient is now hyperopic with a cornea that is too flat. Since tissue cannot be added to the cornea to increase the curvature back to PLANO, an "enhancement" would require the surgeon to laser around the perimeter of the cornea to induce curvature. This is a more difficult correction with a lower likelihood of success, in part because each individual has a limited amount of corneal tissue that can be removed. I assert that an initial overcorrection by the surgeon results in lower expected net benefits to the customer of an enhancement (i.e., warranty work). Thus, H1 predicts that the negative relation between DURATION and quality, as measured by deviations from the PLANO standard, will be weaker for initial procedures that result in an overcorrection (i.e., POST EQU > PLANO) relative to the outcome of an undercorrection (i.e., POST EQU < PLANO).
I test this hypothesis using a maximum likelihood estimation of procedure duration, DURATION, as a function of deviations from the PLANO standard and a term interacting this variable with the overcorrection indicator variable. This estimation again assumes a Weibull distribution for DURATION and adjusts for right censoring in the data.
The results, presented in Table 4, indicate that the magnitude of deviations from the target outcome of PLANO is negatively associated with DURATION for undercorrections (coefficient of -0.439, p-value < .001). That is, for actual outcomes less than PLANO, the further the actual outcome is from the intended outcome of PLANO, the shorter the duration of the initial procedure (i.e., warranty work is more likely). However, the negative relation between deviations from PIANO and DURATION disappears for overcorrections as evidenced by the significantly positive interaction term (p-value < .001). In fact, the sum of the two coefficient estimates is significantly positive (p-value < .001), indicating that larger overcorrections result in longer procedure durations (i.e., increasingly unlikely that a enhancement will be given). Thus, the evidence is consistent with H2, which predicts that a decrease in the expected net benefits to the customer of warranty work weakens the negative relation between quality and future warranty costs.
The results described above should not be interpreted to mean that overcorrections are costless and should, therefore, become the objective. Overcorrections, although not associated with increased warranty costs, may well be very costly for the firm in terms of customer dissatisfaction and reputation damage. A limitation of this study is the lack of data to proxy for these costs (e.g., customer satisfaction data) and the necessary restriction of the analysis to actual warranty costs. Still, the analysis illustrates that the sensitivity of future warranty costs to a quality measure may by moderated by customers' cost-benefit assessments of the warranty work. The extent to which this generalizes to other settings is a question for future research.
Functional Forms of the Quality Cost Function
The final measurement test is of the functional form of the quality (i.e., warranty) cost function. The empirical tests in this section involve estimating and comparing the fits of three models, distinguished only by the three functional form assumptions: the dichotomous form representing the traditional tolerance limit approach, and two forms (linear and quadratic) representing variations of the Taguchi et al. (1989) cost function theory.
Recall that the preceding analysis reveals an asymmetry in the warranty cost function, namely, that warranty costs are not increasing in the magnitude of overcorrections. I, therefore, restrict my functional form tests to the subsample of undercorrected procedures.
To distinguish between the three proposed alternative functional forms of the cost function, I construct three measures of quality, all of which are based on deviations of actual surgery outcomes, POST EQU, from the absolute standard of PLANO (EQU = 0). The first measure, OUTTOL, is constructed to model the traditional tolerance limit approach that assumes quality costs are zero for outcomes within a tolerance limit around the standard. The tolerance limit is arbitrarily chosen as 1 diopter. The second and third measures reflect the theory that any deviation from the outcome standard is costly (Taguchi et al. 1989). These measures, DEVIATION and [DEVIATION.sup.2], assume a linear and quadratic form, respectively. The three measures are:
(1) OUTTOL, representing the traditional tolerance limit cost function (Figure 3, Panel A) and defined as a dichotomous variable equal to 1 if Abs(POST EQU) > 1.0 diopter, and 0 otherwise;
[FIGURE 3 OMITTED]
(2) DEVIATION, representing the linear cost function (Figure 3, Panel B) defined as Abs(POST EQU); and
(3) [DEVIATION.sup.2], representing the quadratic cost function (Figure 3, Panel C) defined as [[Abs(POST EQU)].sup.2]
where POST EQU is the spherical equivalent measure following the procedure.
Three models are estimated in which quality cost (proxied by DURATION) is modeled separately as a function of each of these variables.
Table 5 presents the results of the maximum likelihood estimation of quality costs, proxied by procedure duration, as a function of each of the three measures (Model 4 includes all three measures). I again assume a Weibull distribution for DURATION and adjust for right censoring.
Consistent with expectations, all three measures are significantly negatively associated with DURATION, indicating that larger deviations from a PLANO outcome result in shorter initial procedure duration (i.e., enhancement more likely). However, in a model with all three measures (Model 4), DEVIATION retains its significance (and sign) while the other two measures do not (p-value > .10). Thus, I am able to discriminate between the three non-nested models and accept the DEVIATION (linear) model over the other two (Darnell 1994).
In summary, I find that, using the optimal outcome of PLANO as the quality standard, neither the dichotomous measure (i.e., representing the tolerance limit approach traditionally used in manufacturing), nor the measure based on squared deviations of actual outcomes from the quality standard (i.e., representing the quadratic functional form) add explanatory power over the linear measure's ability to predict quality costs.
Summary of Measurement Improvements
Evidence provided in the tests described above suggests that the standard against which a quality measure is constructed may affect the measure's ability to predict future financial performance. The test of H1 documents that the absolute, optimal outcome of PLANO is the appropriate standard for measuring quality in the current setting. That is, a measure constructed relative to this standard has a stronger association with future warranty costs than a quality measure using the expected outcome as the standard. Hypothesis 2 suggests an asymmetry in the relation between the nonfinancial performance measure and future warranty costs; namely, the strength of the quality-cost relation is increasing in the customer's expected net benefits of warranty work. Finally, the previous section confirms that all deviations from the standard are costly for the firm, consistent with Taguchi et al. (1989). Moreover, a linear quality measure dominates the quadratic measure suggesting a linear quality (warranty) cost function. These findings, taken together, suggest the following "improved" quality measure:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where POST EQU is the spherical equivalent measure following the procedure. Descriptive statistics for this variable are reported in Table 1. Admittedly, the specific findings presented in the preceding three subsections are not generalizable beyond the current setting. However, three measurement issues have been identified--(1) performance measure standard, (2) assumed symmetry of the cost function, and (3) assumed functional form of the cost function--and shown to affect the strength of the relation between the non financial quality measure and future costs. These measurement issues and the theories presented to suggest the measurement alternatives are likely to generalize beyond the measurement of quality and beyond the current setting.
V. QUALITY MANAGEMENT ILLUSTRATION
Implications for Decision Making
Forward-looking quality measures such as the one constructed above play an important role in understanding the relation between quality management decisions and future quality costs (see Figure 4). Management makes numerous quality-related management decisions. Often, there is a significant time lag between these decisions and the incurrence of quality costs such as warranty costs. As a result, a quality measure constructed contemporaneous to the service delivery and with forward-looking properties can play an important role as a connecting link between management decisions and future quality costs.
[FIGURE 4 OMITTED]
Improving the measurement of quality and of the quality cost function (such as was done in the preceding section) is the first important role that management accounting can play in quality management. The second is the linking of quality back to the management decisions that drive quality (and, ultimately, quality costs). A "quality management function" that empirically ties management decisions to a quality measure with forward-looking properties provides timely management feedback for improved decision making and control.
As an illustration, a quality management function (with the "improved" quality measure, DEVIATION, as the dependent variable) is estimated for the current setting. Seven quality-related decisions (i.e., quality drivers) are identified and represent three categories of quality management decisions likely to be important in service settings (see Table 6 for examples). The first category, marketing decisions, include the gender and age of the patient and the magnitudes of required spherical and cylindrical corrections. These variables, and especially the latter two, represent service complexity. Prior accounting research finds product and process complexity to be significant drivers of costs--including quality costs--in manufacturing (Anderson 1993, 1995; Banker and Johnston 1993; Datar et al. 1993; Ittner and MacDuffie 1995; MacArthur and Stranahan 1998) and in healthcare settings (Balakrishnan et al. 1996). Managers of service firms must choose the complexity level of services to provide, leading to the selection of these variables as important quality-related marketing decisions. (7)
Personnel decisions also affect quality and include the experience of the employee delivering the service (e.g., surgeon experience). Learning occurs with experience (or over time) and leads to improved quality (e.g., Fine 1986; Ittner et al. 2001). The importance of personnel decisions will likely be increasing in the extent of interaction between the customer and the employee.
Finally, firms must make resource allocation decisions such as investments in training and equipment maintenance. Prior research documents significant relations between investments in conformance activities and quality (nonconformance) costs in manufacturing firms (e.g., Datar et al. 1993; Ittner 1996; Ittner et al. 2001). (8)
Quality Management Function
Table 7, Model 1 presents the results of an OLS estimation of DEVIATION as a function of the seven quality-related management choice variables. Two control variables are included, an indicator variable for Center A (to control for differences in this center not captured by the other variables) and an indicator variable for overcorrections. (9)
Marketing Decisions
The results show significant relations between the various quality management choice variables and the forward-looking quality measure DEVIATION. Recalling that higher values of DEVIATION correspond to lower quality, the results show that quality is significantly associated with two patient characteristics: age and gender. Controlling for correction magnitudes, quality is decreasing (DEVIATION is increasing) in patient age (coefficient of 0.004, p-value < .001). Quality is also slightly better for men (coefficient of -0.036, p-value < .001) than for women. Finally, the quality measure is decreasing in service complexity as measured by the magnitude of the required spherical correction, MAG SPH. No statistically significant association is found between the quality measure and the magnitude of the cylindrical correction, MAG CYL.
The previous findings are potentially useful for marketing and advertising resource allocation decisions. For example, since quality is higher (and, hence, future warranty costs are lower) for men, relatively more resources can be directed toward advertising that appeals to men. Moreover, pricing can be adjusted to reflect the increased warranty costs associated with larger correction magnitudes. Indeed, firms in this industry have been moving toward discriminatory pricing where higher fees are charged for larger corrections.
Personnel Decisions
Consistent with prior research (e.g., Ittner et al. 2001), quality is increasing (DEVIATION is decreasing) in surgeon experience. (10) That is, more experienced surgeons achieve outcomes closer to PLANO. The coefficient on In EXPERIENCE is -0.077 (p-value < .001). Thus, the experience effect is greatest for low values of EXPERIENCE--at the minimum EXPERIENCE of 11 procedures, every additional ten procedures improves the mean outcome by about 0.07 diopters--and diminishes at higher levels of experience (an improvement of 0.001 diopters for an additional ten procedures at the mean level of experience, and virtually no effect of experience at the maximum experience). (11)
Findings related to personnel decisions have direct implications for operational and management control. With evidence that experience improves quality, managers can focus on ways to move surgeons along the learning curve more quickly or to compensate for lack of experience. As an example, the current firm has developed technology and procedures to monitor outcomes for individual surgeons and to program into the equipment precise adjustments to compensate for differences arising from experience levels (e.g., differences in speed). Moreover, although firms cannot, by law, compensate physicians based on outcomes, the development of appropriately constructed forward-looking performance measures can be used for contracting in non-healthcare settings as a means of directing managerial attention to the long-term (Dikolli 2001).
Resource Allocation Decisions
Evidence that quality is increasing in investments in training and maintenance is also consistent with prior research. Specifically, deviations from the optimal outcome of PLANO are lower following the firm-wide training effort and for centers with higher per-procedure expenditures on maintenance. These results have direct uses for evaluating past investments and predicting the returns on future, similar ones. Moreover, such evaluations are made more timely with the use of contemporaneously determined forward-looking performance measures.
Comparison of Alternative Measures
Models 2 and 3 in Table 7 present results of the estimation of the quality management function using the two alternative measures, [DEVIATION.sup.2] and OUTTOL, respectively, as the quality outcome measure to be managed (i.e., the dependent variable). Inferences regarding quality management decisions are qualitatively similar as those for the quality outcome measure DEVIATION, with one important exception--namely, when using the measure shown to be a better predictor of future warranty costs (i.e., the more goal-congruent measure DEVIATION) as the quality outcome variable to be managed, improvements in quality following the training program are revealed. On the other hand, if managers in this setting were to use either of the other two quality measures proposed and evaluated in the previous section (i.e., [DEVIATION.sup.2] or OUTTOL), then they would incorrectly infer that no future resources should be allocated to training since training appears to have no relation to the quality measure used. Thus, even with subtle measurement differences such as those that distinguish DEVIATION from the other two measures, important differences can arise in managerial decisions.
In summary, this illustration shows that the forward-looking quality measure is associated with quality management decisions likely to be important in the current setting. Thus, the estimation of a quality management function provides a tool for managers to evaluate past (and future) decisions based on their likely effect on quality and future warranty costs. The value of this tool lies in the measure of quality that is 'contemporaneous to the service delivery (and, therefore, more timely with respect to decisions) and has forward-looking properties.
VI. SUMMARY AND CONCLUSIONS
Prior research documents that nonfinancial performance measures are often leading indicators of financial performance. In this paper, I examine the effects of measurement alternatives on the forward-looking properties of a quality performance measure. Specifically, the quality and quality cost function measurement issues examined are: (1) the alternative standards against which quality is measured, (2) whether variations in quality have a uniform affect on future quality costs (i.e., consideration of an asymmetric relation between quality and future quality costs), and (3) the exact functional form or the relation between a quality measure and future quality costs. Research from operations, management, and marketing is used to develop hypotheses regarding the effect of these measurement alternatives on the strength of the forward-looking properties of the quality measure.
Using proprietary data from a medical services firm, I find, first, that the standard against which quality is measured affects the strength of the association between the quality measure and future quality-related warranty costs and, hence, the measure's role as a forward-looking performance indicator. Although marketing theory supports the use of the customer's ex ante expectation as the standard against which quality should be measured (e.g., Cronin and Taylor 1992, 1994; Parasuraman et al. 1985; Patterson 1993; Stank et al. 1999), an absolute, optimal outcome standard dominates an expected outcome standard in the current setting. Next, the warranty cost function is shown empirically to be asymmetric in the current setting. This result is consistent with the generalizable hypothesis that the strength of the relation between quality and future quality-related warrant costs is a function of customers' expected net benefit of the warranty work. Finally, alternative functional forms of the warranty cost function as suggested by current manufacturing practice and operations theory are considered. The empirical results do not support the tolerance limit approach that is commonly used in manufacturing settings and that assumes small variations from optimal quality are not costly for the firm. Instead, the results suggest that all deviations from optimal quality are costly in the current setting, consistent with quality theory in operations research (e.g., Taguchi et al. 1989). Moreover, the data support a linear (as opposed to an exponential) relation between the quality measure and future quality costs in the current setting.
To highlight the implications of this study for management accounting, the paper concludes with an illustration of how an "improved" forward-looking performance measure facilitates management decision making by capturing the effects of those decisions on future costs. For example, findings related to the association between the quality measure and expenditures on various conformance efforts can direct a firm to the optimal allocation of resources to those efforts (e.g., training seems to be more effective than the hiring of more skilled surgeons). Moreover, identifying the relation between various marketing decisions and the quality measure may suggest future advertising or pricing strategies. The fact that quality is measured contemporaneous to the service delivery and has forward-looking properties further strengthens its usefulness in evaluating past (and future) quality-related decisions in a timely manner. Finally, the illustration shows that decisions can differ when alternative forms of the quality measure are used.
As is characteristic of field research, the specific findings of this study are limited in their generalizability. However, the evidence provided is consistent with the very general proposition that the construction of a quality measure--that is, the selection of measurement alternatives--may improve its ability to predict future financial performance and, hence, its usefulness for decision making and control. Moreover, the measurement issues identified and the theories presented to suggest measurement alternatives arguably generalize beyond the measurement of quality and beyond the current setting. Finally, this paper speaks to a void in the management accounting research related to the factors that affect the strength of leading indicator relations. Additional theoretical and empirical identification of factors that affect leading indicator relations (e.g., customer satisfaction and future financial performance, Ittner and Larcker 1998a) remains an open and promising area for future research.
TABLE 1
Descriptive Statistics
Panel A: Descriptive Statistics (a)
No. of Center-Months 76
No. of Initial Procedures 4,470
No. of Enhancements (b) 243
Enhancement Rate (c) 5.44%
Procedure-Level Variables Mean Std. Dev. Min. Max.
DURATION (with censored
observations) 265.52 148.63 11 546
DURATION (without censored
observations) 255.48 71.91 118 491
DEVIATION (quality measure) 0.26 0.45 0 4.38
MAG SPH (magnitude
of the SPH correction) 4.63 2.62 0.00 18.00
MAG CYL (magnitude of
the CYL correction) 1.06 0.91 0.00 6.00
AGE 38.20 8.98 17 68
GENDER (1 = male) 0.50 0.50 0 1
PRE EQU -5.16 2.69 -19.25 -0.63
POST EQU -0.08 0.66 -4.38 4.63
Monthly-level variables
EXPERIENCE (d) 943.53 1,448.00 11 5,102
TRAINING (1 = post-July 97) 0.13 0.34 0 1
Avg MAINT 15.17 20.53 2.31 51.62
VOLUME 81.97 88.91 6 309
Panel B: Descriptive Statistics by Center (e)
Center: A B C
No. of Months 18 15 14
No. of Initial Procedures 3,263 290 393
No. of Enhancements (b) 202 3 25
Enhancement Rate (c) 6.19% 1.03% 6.36%
Procedure-Level Variables
DURATION 299.02 190.15 167.01
(with censored observations) (142.47) (136.81) (109.02)
DURATION 265.00 301.00 196.20
(without censored observations) (68.96) (122.65) (68.90)
DEVIATION 0.21 0.41 0.43
(quality measure) (0.39) (0.52) (0.66)
MAG SPH 4.88 3.42 4.32
(magnitude of the SPH correction) (2.74) (1.78) (2.32)
MAG CYL 1.15 0.66 0.93
(magnitude of the CYL correction) (0.92) (0.68) (0.83)
AGE 38.28 36.47 38.21
(8.98) (8.51) (9.02)
GENDER (1 = male) 0.50 0.47 0.49
(0.50) (0.50) (0.50)
PRE EQU -5.46 -3.75 -4.77
(2.80) (1.81) (2.36)
POST EQU 0.04 -0.41 -0.43
(0.66) (0.52) (0.66)
Monthly-level variables
EXPERIENCE (d) 3,275.28 221.87 262.50
(1,267.47) (157.21) (197.89)
TRAINING (1 = post-July 97) 0.11 0.13 0.14
(0.32) (0.35) (0.36)
Avg MAINT 51.62 2.91 2.31
(0.00) (0.00) (0.00)
VOLUME 230.50 34.60 45.57
(48.21) (26.29) (26.56)
All
Center: D E Centers
No. of Months 13 16 76
No. of Initial Procedures 186 338 4,470
No. of Enhancements (b) 4 9 243
Enhancement Rate (c) 2.15% 2.66% 5.44%
Procedure-Level Variables
DURATION 119.56 201.60 265.52
(with censored observations) (105.96) (132.26) (148.63)
DURATION 213.50 209.89 255.48
(without censored observations) (36.82) (34.34) (71.91)
DEVIATION 0.28 0.43 0.26
(quality measure) (0.43) (0.51) (0.45)
MAG SPH 3.51 4.23 4.63
(magnitude of the SPH correction) (2.04) (2.08) (2.62)
MAG CYL 0.70 0.91 1.06
(magnitude of the CYL correction) (0.85) (0.87) (0.91)
AGE 38.95 38.42 38.20
(8.65) (9.43) (8.98)
GENDER (1 = male) 0.57 0.49 0.50
(0.50) (0.50) (0.50)
PRE EQU -3.86 -4.69 -5.16
(1.98) (2.09) (2.69)
POST EQU -0.28 -0.43 -0.08
(0.43) (0.51) (0.66)
Monthly-level variables
EXPERIENCE (d) 120.15 261.75 943.53
(95.14) (191.95) (1,448.00)
TRAINING (1 = post-July 97) 0.15 0.13 0.13
(0.38) (0.34) (0.34)
Avg MAINT 7.88 2.81 15.17
(0.00) (0.00) (20.53)
VOLUME 24.23 38.06 81.97
(16.55) (22.79) (88.91)
(a) For the quality measure, DEVIATION, the quality cost proxy
measure, DURATION, and the procedure-level variables (MAG SPH,
MAG CYL, AGE, GENDER, PRE EQU, and POST EQU), the statistics
are computed over all outcomes. For the monthly-level variables
(VOLUME, TRAINING, Avg MAINT, and EXPERIENCE), the statistics
are computed over all center-months.
(b) The number of initial procedures in the database that
required an enhancement during the test period.
(c) Enhancement rate is the percentage of initial procedures that
required at least one enhancement. Sixteen patients required two
enhancements. Note that this is a right-censored statistic.
(d) The experience variable is computed using an additional 11
months of outcomes data for which there was no corresponding
monthly data.
(e) Standard deviations in parentheses.
Variable Definitions:
DURATION = the number of days between the initial procedure and
the date of the enhancement (or to the end of the test
period for censored observations). This variable
proxies for future quality costs;
DEVIATION = the difference between the actual spherical equivalent
outcome and an outcome of PLANO (i.e., spherical
equivalent equal to zero) for observations in which
POST EQU < 0 and zero otherwise. This variable is the
contemporaneous nonfinancial quality measure;
MAG SPH = the magnitude of the spherical equivalent correction
needed (i.e., the severity of the myopia);
MAG CYL = the magnitude of the cylindrical correction needed
(i.e., the severity of the astigmatism);
AGE = patient age;
GENDER = gender of patient (1 = male);
PRE EQU = the patient's spherical equivalent measure (in diopters)
just prior to the procedure;
POST EQU = the patient's spherical equivalent measure (in diopters)
after the procedure;
EXPERIENCE = the cumulative number of procedures;
TRAINING = indicator variable identifying the center-months
following the firm-wide training program (1 =
post-July 97);
Avg MAINT = maintenance expenditure per procedure for the center; and
VOLUME = the number of procedures performed in a center-month.
TABLE 2
Correlations
Panel A: Procedure-Level Variables (a)
Spearman Rank Correlations
DURATION DEVIATION MAG SPH MAG CYL
DURATION 0.004 0.087 * 0.082 *
DEVIATION 0.011 -0.007 -0.065 *
MAG SPH 0.081 * 0.018 0.068 *
MAG CYL 0.079 * -0.040 * 0.057 *
AGE 0.005 0.087 * 0.044 * 0.095 *
GENDER 0.017 -0.062 * -0.083 * -0.008
PRE EQU -0.093 * -0.011 -0.986 * -0.224 *
POST EQU 0.058 * -0.838 * 0.100 * 0.080 *
Pearson Correlations
Spearman Rank Correlations
AGE GENDER PRE EQU POST EQU
DURATION 0.003 0.018 -0.100 * 0.071 *
DEVIATION 0.056 * -0.057 * 0.015 -0.924 *
MAG SPH 0.049 * -0.096 * -0.981 * 0.062 *
MAG CYL 0.132 * -0.014 -0.234 * 0.096 *
AGE -0.021 -0.069 * -0.028
GENDER -0.021 0.099 * 0.048 *
PRE EQU -0.059 * 0.083 * -0.075 *
POST EQU -0.038 0.045 * -0.111 *
Pearson Correlations
Panel B: Monthly Level Variables (a)
Spearman Rank Correlations
EXPERIENCE ln EXPERIENCE TRAINING Avg MAINT
EXPERIENCE 1.000 * 0.319 * 0.431 *
ln EXPERIENCE 0.858 * 0.319 * 0.431 *
TRAINING 0.125 0.231 -0.018
Avg MAINT 0.896 * 0.810 * -0.032
Pearson Correlations
* Indicates significance at the 0.01 level.
(a) Both Pearson (lower diagonal) and Spearman (upper diagonal)
correlations are provided.
The variables are defined in Table 1.
TABLE 3
Comparison of the Optimal and Expectation-Adjusted
Quality Standards (a)
Dependent Variable: DURATION
Independent Variable Model 1 Model 2 Model 3
Intercept 7.815 *** 7.701 *** 7.829 ***
Deviation from PLANO -0.500 *** -0.428 ***
Deviation from E[outcome] -0.497 *** -0.151
[gamma] (scale parameter) 0.638 0.669 0.632
Pseudo-[R.sup.2] 6.63% 2.31% 6.15%
LR Statistics:
vs. intercept only model 18.42 *** 7.72 *** 19.22 ***
No. of noncensored values 36
No. of right censored values 277
*** Indicates significance at the .001 level (two-tailed test).
(a) This table presents the results of a maximum likelihood estimation
of procedure DURATION (i.e., time to enhancement) as a function of
quality as measured against one of two standards, the optimal outcome
of PLANO (i.e., the absolute quality measure) or the expected outcome
(i.e., the expectation-adjusted quality measure). The model is
estimated for the subsample of observations in which POST EQU < 0 and
E[outcome] [not equal to] 0 (PLANO). The estimation assumes a Weibull
distribution for duration and adjusts for right censoring for
procedures not requiring an enhancement by the end of the test period.
Coefficient estimates are shown along with estimates of the Weibull
distribution scale parameter, [gamma].
Two fit statistics, the pseudo-[R.sup.2] and the likelihood ratio
statistic, are also presented.
Variable Definitions:
DURATION = the number of days between the initial procedure and the
date of the enhancement (or to the end of the test period
for censored observations);
PLANO = the optimal spherical equivalent outcome; represents the
optimal quality standard and assumes the value of 0;
E[outcome] = the expected spherical equivalent outcome identified by
surgeon just prior to the procedure; represents the
expectation quality standard; and
POST EQU = the actual spherical equivalent measure after the
procedure.
Deviation from PLANO (i.e., Absolute Quality Measure) = Abs(POST EQU
- PLANO) Deviation from E[outcome] (i.e., Expectation-Adjusted Quality
Measure) = Abs(POST EQU - E[outcome])
TABLE 4
Test for Asymmetry in the Quality Cost Function (a)
Dependent
Independent Variable Variable: DURATION
Intercept 7.574 ***
Deviation from PLANO -0.439 ***
Deviation from PLANO * OVER 0.835 ***
[gamma] (scale parameter) 0.555
Pseudo-[R.sup.2] 5.45%
LR Statistics:
vs. intercept only model 122.24 ***
vs. model without interaction 63.62 ***
No. of noncensored values 280
No. of right censored values 4,538
No. of undercorrections (mean absolute
deviation) 2,400 (0.563)
No. of overcorrections (mean absolute
deviation) 2,418 (0.572)
*** Indicates significance at the .001 level (two-tailed tests).
(a) This table presents the results of a maximum likelihood estimation
of procedure DURATION (i.e., time to enhancement) as a function of
quality as measured against the optimal (PLANO) standard and an
interaction of this variable and an overcorrection indicator variable,
OVER. The estimation assumes a Weibull distribution for duration and
adjusts for right censoring for procedures not requiring an enhancement
by the end of the test period.
Coefficient estimates are shown along with estimates of the Weibull
distribution scale parameter, [gamma].
Two fit statistics, the pseudo-[R.sup.2] and the likelihood ratio
statistic, are also presented.
Variable Definitions:
DURATION = the number of days between the initial procedure and the
date of the enhancement (or to the end of the test period
for censored observations);
PLANO = the optimal spherical equivalent outcome; represents
the optimal quality standard and assumes the value of 0;
POST EQU = the actual spherical equivalent measure after the procedure;
and
OVER = an indicator variable equal to one for overcorrections.
Deviation from PLANO standard (i.e., Absolute Quality Measure) =
Abs(POST EQU - PLANO)
TABLE 5
Test of Alternative Functional Forms of the Quality Cost Function (a)
Dependent Variable: DURATION
Model 1 Model 2 Model 3
Dichoto- Linear Quadratic
mous Model Model Model Model 4
Independent
Variable
Intercept 7.305 *** 7.465 *** 7.295 *** 7.505 ***
OUTTOL -0.371 *** 0.226
DEVIATION -0.355 *** -0.487 ***
[DEVIATION.sup.2] -0.093 *** 0.008
[gamma] (scale
parameter) 0.569 0.565 0.562 0.565
Pseudo-[R.sup.2] 0.82% 2.13% 1.65% 1.99%
LR Statistics:
vs. intercept
only model 13.92 *** 32.95 *** 25.92 *** 34.86 ***
vs. Model 2 1.91
No. of noncensored
values 211
No. of right
censored values 2,165
*** Indicates significance at the .001 level (two-tailed tests).
(a) This table presents the results of a maximum likelihood
estimation of procedure DURATION (i.e., time to enhancement)
as a function of one of the three quality measures: OUTFOL,
DEVIATION, or [DEVIATION.sup.2] (see Figure 3). The model is
estimated for the subsample of observations in which POST EQU
< 0. The estimation assumes a Weibull distribution for duration
and adjusts for right censoring for procedures not requiring an
enhancement by the end of the test period.
Coefficient estimates are shown along with estimates of the
Weibull distribution scale parameter, [gamma].
Two fit statistics, the pseudo-[R.sup.2] and the likelihood ratio
statistic, are also presented.
Variable Definitions:
DURATION = the number of days between the initial procedure
and the date of the enhancement (or to the end
of the test period for censored observations);
OUTTOL = represents the traditional tolerance limit cost
function (Figure 3, Panel A); defined as a
dichotomous variable equal to 1 if Abs(POST EQU)
> 1.0 diopter, and 0 otherwise;
DEVIATION = represents the linear cost function (Figure 3,
Panel B); defined as Abs(POST EQU); and
[DEVIATION.sup.2] = represents the quadratic cost function (Figure 3,
Panel C) defined as [[Abs(POST EQU)].sup.2].
TABLE 6
Quality-Related Management Decisions in Service Firms
Decision Examined in
Type of Decision Current Study
Marketing Decisions
* complexity of service [right arrow] magnitudes of spherical and
delivered cylindrical corrections
* heterogeneity of
services delivered
(i.e., service mix)
* location of service
delivery
* other customer- [right arrow] gender and age of the
specific characteris- patient
tics (individual or
firm)
Personnel Decisions
* hiring--prior [right arrow] surgeon experience
experience
* hiring--level and
quality of prior
training/education
* hiring--level of risk
aversion
* incentive system
design
* performance
measurement system
design
Resource Allocation
Decisions
* investments in [right arrow] investments in surgeon
training and technician training
* investments in [right arrow] investments in laser
maintenance equipment maintenance
* investments in
appraisal activities
* firm-wide quality
initiatives (e.g.,
TQM)
TABLE 7
Estimation of the Quality Management Function (a)
Model 2
Model 1 [DEVIATION. Model 3
Dependent Variable: DEVIATION sup.2] OUTTOL
Exp.
Independent Variable Sign
Intercept -0.298 * -1.150 *** -7.458 ***
Marketing Decisions
MAG SPH + 0.001 *** 0.016 *** 0.085 ***
MAG CYL + -0.004 -0.006 0.037
AGE ? 0.004 *** 0.008 *** 0.034 ***
GENDER (1 = male) ? -0.036 *** -0.064 *** -0.322 **
Personnel Decisions
ln EXPERIENCE -- -0.077 *** -0.134 *** -0.112 ***
Resource Allocation
Decisions
TRAINING -- -0.063 *** -0.044 -0.432
Avg MAINT -- -0.019 *** -0.045 *** -0.149 **
CENTER A (control) 1.085 *** 2.389 *** 7.948 **
OVER (control) -0.402 *** -0.404 *** -15.529
Adjusted [R.sup.2] 23.69% 8.60%
F-statistic 167.68 *** 51.51 ***
Pseudo-[R.sup.2] 18.12%
LR Statistic 393.60 ***
*, **, *** Indicates significance at the .05, .01, .001 levels,
respectively (two-tailed tests).
(a) This table presents results of the estimation of the quality
management function (n = 4,833). Models 1 and 2 are OLS estimations
of DEVIATION and [DEVIATION.sup.2], respectively, modeled as a
function of the quality-related decisions likely to be important
in the current setting, and two control variables (an indicator for
CENTER A and an indicator for overcorrections, OVER). Model 3 is a
maximum likelihood (logistic) estimation of the OUTTOL quality
measure as a function of the decision variables. Higher values of
the DEVIATION, [DEVIATION.sup.2], and OUTTOL variables indicate
lower quality. Higher values of MAG SPH and MAG CYL indicate
higher service complexity. For ease of interpretation, all variables
except the indicator variables are mean-centered.
Variable Definitions:
MAG SPH = the magnitude of the spherical equivalent correction
needed (i.e., the severity of the myopia);
MAG CYL = the magnitude of the cylindrical correction needed
(i.e., the severity of the astigmatism);
AGE = age of patient;
GENDER = gender of patient (1 = male);
ln EXPERIENCE = the log of the cumulative number of procedures
performed by the surgeon;
TRAINING = indicator variable identifying the center-months
following the firm-wide training program
(1 = post-July 97);
Avg MAINT = mean maintenance expenditure per procedure for the
center;
CENTER A = an indicator variable equal to 1 for Center A
observations; and
OVER = an indicator variable equal to 1 for overcorrections.
(1) Taguchi et al. (1989) argue that important sources of cost are the "hidden" losses of reputation damage and lost customer sales. They propose a quadratic form to account for these hidden losses.
(2) The original data consisted of ten centers. I omitted five centers because they entered the sample late in the test period and/or they had too few procedures during the test period. I allowed a center to enter the sample only when it reached a volume exceeding ten procedures in a month.
(3) During the period of study, there were no excimer lasers in the U.S. with FDA approval for the correction of hyperopia (i.e., spherical equivalent > 0). Moreover, no enhancements for EQU > 1.75 were performed in the Canadian centers. Since this externally imposed restriction would reduce the power of the statistical tests (or possibly bias the results), all observations for U.S. centers with POST EQU > 0 and Canadian centers with POST EQU > 1.75 were eliminated from the analysis.
(4) Although I do not have data to corroborate this claim, it also seems reasonable to assume that hidden quality costs (Taguchi et al. 1989) related to customer dissatisfaction and reputation damage are also decreasing in elapsed time.
(5) Modeling the duration of the initial procedure also offers a significant econometric advantage. Because enhancements occur, on average, over eight months following the initial procedure, and because my sample includes a fairly short time series, the data are significantly right-censored. A duration modeling approach adjusts for right censoring, allowing the information in both censored and uncensored observations to be used in a way that provides consistent estimates of the parameters. In fact, this is one of the most significant advantages of the survival analysis technique (Allison 1995).
(6) The Weibull distribution allows for either a monotonically increasing or decreasing hazard function (Greene 1997) and is less restrictive than other nonnegative distributions often used in duration models (e.g., exponential or lognormal distributions). With this specification, each coefficient estimate represents the derivative of the log of E[duration] with respect to a unit change in the associated independent variable (i.e., [delta] In E[DURATION]/ [delta][x.sub.h] = [[beta].sub.h]). The estimated Weibull distribution scale parameter, [gamma], indicates the duration dependence of the data.
(7) In some settings, the firm has a large degree of control over which customers will receive services (e.g., healthcare settings in which the physician can "screen" patients for a particular treatment). In other settings, although the degree of control is not as great, the firm influences customer selection by its pricing and advertising decisions.
(8) The traditional view of cost of quality management assumes that nonconformance costs can be reduced only by increasing conformance costs (Feigenbaum 1983; Ireson 1971). More recent theories suggest that, in a dynamic setting, both conformance and nonconformance costs can be reduced with organizational learning (Crosby 1979; Fine 1986; Love et al. 1995) and with capital investments in quality initiatives that allow for continual quality improvement. Although empirical evidence supports the dynamic model (e.g., Foster and Adam 1996; Ittner 1996; Ittner et al. 2001), the current study does not capture this dynamic environment.
(9) Equivalent results are obtained for a subsample of undercorrections.
(10) Note that the model was estimated using the natural log of experience (measured as the cumulative number of procedures by the surgeon) to allow for diminishing returns. Although not the focus of this section, this illustrates another opportunity for management accountants, namely, to investigate ways to improve the measurement of the quality management function.
(11) That is, [delta] DEVIATION/[delta] EXPERIENCE = -0.007 * (1/EXPERIENCE). At the minimum EXPERIENCE level of 11 procedures, the improvement in DEVIATION associated with an additional ten procedures is -0.077 * (1/ 11) * 10 = 0.07 diopters. Similarly, at the mean EXPERIENCE level of 944 procedures, the improvement is -0.077 * (1/944) * 10 = 0.001 diopters. A rough rule of thumb is 1 diopter for every 100 (to 150) feet of visual acuity (e.g., 20/20 vs. 20/120).
REFERENCES
Albright, T. L., and H. P. Roth. 1992. The measurement of quality costs: An alternative paradigm. Accounting Horizons 6 (2): 15-27.
Allison, P. D. 1995. Survival Analysis Using the SAS System: A Practical Guide. Cary, NC: SAS Institute Inc.
Anderson, E. W., and M. W. Sullivan. 1993. The antecedents and consequences of customer satisfaction for firms. Marketing Science 12 (2): 125-143.
--, C. Fornell, and D. R. Lehmann. 1994. Customer satisfaction, market share, and profitability: Findings from Sweden. Journal of Marketing 58 (3) 53-66.
Anderson, S. W. 1993. Measuring manufacturing flexibility: The impact of product mix complexity on operating performance and manufacturing overhead cost. Ph.D. dissertation, Harvard University.
--. 1995. Measuring the impact of product mix heterogeneity on manufacturing overhead cost. The Accounting Review 70 (3): 363-387.
Apte, U. M., U. S. Karmarkar, and R. Pitbladdo. 1997. Quality management in services: Analysis and measurement. In The Practice of Quality Management, edited by P. J. Lederer, and U. S. Karmarkar. Boston, MA: Kluwer Academic Publishers.
Balakrishnan, R., T. Gruca, and D. Nath. 1996. The effect of service capability on operating costs: An empirical study of Ontario hospitals. Contemporary Accounting Research 13 (1): 177-206.
Banker, R., and H. H. Johnston. 1993. An empirical study of cost drivers in the U.S. airline industry. The Accounting Review 68 (3): 576-601.
--, G. Potter, and D. Srinivasan. 2000. An empirical investigation of an incentive plan that includes nonfinancial performance measures. The Accounting Review 75 (1): 65-92.
Bearden, W. O., and J. E. Teel. 1983. Selected determinants of consumer satisfaction and complaint reports. Journal of Marketing Research 20 (1): 21-28.
Cronin, J. J., and S. A. Taylor. 1992. Measuring service quality: A reexamination and extension. Journal of Marketing 56 (3): 55-68.
--, and --. 1994. Modeling patient satisfaction and service quality. Marketing Health Services 14 (1): 34-44.
Crosby, P. B. 1979. Quality is Free. New York, NY: McGraw-Hill.
Datar, S. M., S. Kekre, T. Mukhopadhyay, and K. Srinivasan. 1993. Simultaneous estimation of cost drivers. The Accounting Review 68 (3): 602-614.
--, S. C. Kulp, and R. A. Lambert. 2001. Balancing performance measures. Journal of Accounting Research 39 (1): 75-92.
Darnell, A. C. 1994. A Dictionary of Econometrics. Lyme, NH: Edward Elgar Publishing, Inc.
Dikolli, S. S. 2001. Agent employment horizons and the contracting demand for leading indicators of future performance. Journal of Accounting Research 39 (3): 481-494.
Feigenbaum, A. V. 1983. Quality costs. In Total Quality Control, 3rd edition. New York, NY: McGraw-Hill.
Feltham, G. A., and J. Xie. 1994. Performance measure congruity and diversity in multi-task principal/ agent relations. The Accounting Review 69 (3): 429-453.
Fine, C. H. 1986. Quality improvement and learning in productive systems. Management Science 32 (10): 1301-1315.
Foster, S. T., and E. E. Adam, Jr. 1996. Examining the impact of speed of quality improvement on quality-related costs. Decision Sciences 27 (4): 623-646.
Greene, W. H. 1997. Limited dependent variable and duration models. In Econometric Analysis, 3rd edition. Upper Saddle River, NJ: Prentice Hall, Inc.
Ireson, W. 1971. Handbook of Industrial Engineering and Management. 2nd edition. Englewood Cliffs, NJ: Prentice Hall Inc.
Ittner, C. D., and J. P. MacDuffie. 1995. Explaining plant-level differences in manufacturing overhead: Structural and executional cost drivers in the world auto industry. Production and Operations Management 4 (4): 312-334.
--. 1996. Exploratory evidence on the behavior of quality costs. Operations Research 44 (1): 114-130.
--, and D. Larcker. 1998a. Are non-financial measures leading indicators of financial performance? An analysis of customer satisfaction. Journal of Accounting Research 36 (Supplement): 1-35.
--, and --. 1998b. Innovations in performance measurement: Trends and research implications. Journal of Management Accounting Research 10: 205-238.
--, V. Nagar, and M. V. Rajan. 2001. Empirical examination of quality learning models. Management Science 47 (4): 563-578.
Kim, M., and W. Liao. 1994. Estimating hidden quality costs with quality loss. Accounting Horizons 8 (1): 8-18.
Kinney, W. R., Jr. 2001. Accounting scholarship: What is uniquely ours? The Accounting Review 76 (2): 275-284.
Kotler, P., and P. N. Bloom. 1984. Marketing Professional Services. Englewood Cliffs, NJ: Prentice Hall.
Lambert, R. A. 1998. Customer satisfaction and future financial performance: Discussion of: Are nonfinancial measures leading indicators of financial performance? An analysis of customer satisfaction. Journal of Accounting Research 36 (Supplement): 37-46.
Love, C. E, R. Guo, and K. H. Irwin. 1995. Acceptable quality level versus zero-defects: Some empirical evidence. Computers & Operations Research 22 (4): 403-417.
MacArthur, J. B., and H. A. Stranahan. 1998. Cost driver analysis in hospitals: A simultaneous equations approach. Journal of Management Accounting Research 10:280-312.
Nagar, V. 1998. The explanatory power of non-financial measures in bank earnings prediction. Working paper, University of Michigan, Ann Arbor, MI.
--, and M. Rajan. 2001. Revenue implications of product quality measures. The Accounting Review 76 (4): 495-513.
Nelson, E. C., R. T. Rust, A. Zahorik, and R. L. Rose. 1992. Do patient perceptions of quality relate to hospital financial results? Journal of Health Care Marketing 12 (4): 6-13.
Oliver, R. L. 1980. A cognitive model of the antecedents and consequences of satisfaction decisions. Journal of Marketing Research 17 (4): 460-469.
Parasuraman, A., V. Zeithaml, and L. Berry. 1985. A conceptual model of service quality and its implications for future research. Journal of Marketing (Fall): 41-50.
--, and --. 1994. Reassessment of expectations as a comparison standard in measuring service quality: Implications for further research. Journal of Marketing 58 (1): 111-124.
Patterson, P. G. 1993. Expectations and product performance as determinants of satisfaction for a high-involvement purchase. Psychology & Marketing 10 (5): 449-465.
Phillips, L. D., D. R. Chang, and R. Buzzell. 1983. Product quality, cost position, and business performance: A test of some key hypotheses. Journal of Marketing 47 (2): 26-43.
Stank, T. P., T. J. Vickery, and K. Shawnee. 1999. Effect of service supplier performance on satisfaction and loyalty of store managers in the fast food industry. Journal of Operations Management 17 (4): 429-447.
Taguchi, G., E. A. Elsayed, and T. Hsiang. 1989. Quality Engineering in Production Systems. NY: McGraw-Hill.
Zeithaml, V. A., L. L. Berry, and A. Parasuraman. 1996. The behavioral consequences of service quality. Journal of Marketing 60 (2): 31-46.
I am grateful for the assistance of my dissertation committee: Shannon Anderson (chair), David Butz, Raffi Indjejikian, Bill Lanen, and Leon Wyszewianski. I also thank the following individuals for their many helpful comments and suggestions: Ramji Balakrishnan, Shane Dikolli, Leslie Eldenburg, Steven Kachelmeier, Bill Kinney, Lisa Koonce, Joan Luft (associate editor), Mary Lea McAnally, Margaret Shackell-Dowell, Naomi Soderstrom, two anonymous reviewers, participants at the 2000 AAA/IMA Management Accounting Research Conference, the 2001 AAA Annual Meeting and workshop participants at The University of Texas at Austin, Stanford University, Rice University, University of North Texas, and The University of Texas at Dallas. I am also grateful to those individuals at my research site for their time and cooperation and for their efforts in providing data. This paper is based on my dissertation completed at the University of Michigan.
Karen L. Sedatole The University of Texas at Austin
In addition, make sure to read these articles:
- Chrysler To Consider Offering Extended Warranty In Europe.
- Lean: A supplier's weapon against warranty costs
- Force10 Networks Adopts SigmaQuest's On-Demand Suite of Performance Intelligence...
- Warranty Sharing Heats Up.
- For technical assistance ... press 3: surveys showcustomers are mad and not taking it...
- Nordyne honored for quality production
- Ford E-Business Will Boost Quality, Slash Cost.
- GM To Use New Technology To Produce Truck Frames.
- Do dealers know their warranty costs?
- Product Warranty Handbook.
- One size does not fit all.
- Vibration-proof fasteners. (Products).
- What shipowners want from ashipyard.
- Time is money: focus on the schedule to keep clients happy and put money in your pocket.
- Unprecedented Demand for Automotive Software Throws Up Tremendous Growth Opportunities and...
Email
Digg It
del.icio.us
Related Resources
