Further evidence on the extent and origins of JIT's profitability effects.

I. INTRODUCTION

Since the early 1980s, many U.S. firms have adopted just-in-time production management (JIT). However, empirical evidence on the association between JIT adoption and financial performance is mixed. Although Young and Selto (1991) concluded from early, site-based evidence

that JIT helped adopters reduce non-value-added costs, more recent empirical studies (e.g., Huson and Nanda 1995; Balakrishnan et al. 1996) have generally found that JIT does not necessarily lead to improved financial performance.

We re-examine the association between JIT adoption and financial performance. Our study most closely resembles Balakrishnan et al. (1996) (hereafter BLV). Using samples of 46 JIT adopters and matched non-adopters, BLV concluded that adopters improved inventory turnover relative to non-adopters. However, BLV did not find that JIT adoption, on average, is associated with improved return on assets (ROA). Instead, BLV's evidence suggests that JIT-related ROA improvement is restricted to adopters with diffuse customer bases, which outperform adopters with concentrated customer bases and non-adopting industry counterparts. BLV inferred that "captive" firms with one or more major customers may not retain JIT's financial benefits, but instead may be forced to pass along such benefits to their more powerful customers. (1)

We extend BLV's study by using a larger and more recent sample to examine the association between JIT adoption and changes in financial performance. We also investigate whether JIT adopters' relative ROA changes are primarily attributable to changes in either the profit margin or asset turnover component of ROA.

We identify JIT adopters and match them with non-adopting firms, comparing changes in their respective financial performances from a three-year pre-adoption period to a three-year post-adoption period. To validate our sample selection, we first examine the relation between JIT adoption and changes in both inventory turnover and inventory-to-total assets ratios. We find that adopters enjoy an inventory turnover increase six to eight times greater than that of non-adopters, representing a 16 to 20 percent increase over pre-adoption turnover levels. JIT adopters also reduce inventory-to-total assets more substantially than do non-adopters.

In contrast to BLV, our results suggest that, on average, JIT adopters improve ROA relative to non-adopters, and that this relation does not vary with the concentration of adopters' customer bases. Moreover, our results indicate that JIT adopters improve both the profit margin and asset turnover components of ROA, relative to their non-adopting counterparts. We then compare the relative contributions of changes in profit margin and asset turnover to JIT adopters' relative ROA changes. We find that adopters' relatively improved ROAs derive largely from improved profit margins, suggesting that JIT's primary benefits stem from the elimination of non-value-adding production costs, rather than from mere reductions in total investment arising from leaner inventories. This finding regarding the source of ROA improvement is consistent with Alles et al. (1995), who suggest that the absence of buffer stocks increases the incentive to eliminate production problems, thereby eliminating the costs associated with such problems.

Two differences between our study and BLV lead to fundamental differences in the studies' JIT adopter samples. First, as an early study examining the relation between financial performance and JIT, BLV used a sample of 46 firms that adopted JIT prior to 1990. In contrast, our sample of 201 adopters includes not only such "early" adopters, but also 78 post-1989 adopters. Second, BLV focused (in part) on customer concentration's effect on the relation between JIT and financial performance; accordingly, BLV limit their sample to firms reporting operations in fewer than four three-digit SICs (Balakrishnan et al. 1996, 189). Our primary interest is the sample-wide association between JIT and financial performance, and we examine the roles of customer concentration and other possible intervening variables in an ex post attempt to refine our understanding of that association. Thus, we employ no ex ante span-of-operations screen.

These differences in research timing and focus contribute to a significant between-study difference in sample size. Our sample of adopters is more than four times larger than BLV's, which increases the power of our tests. Moreover, BLV's span-of-operations screen, and our omission of such a screen, contributes to a substantial difference in average JIT adopter size between the two studies--our adopters are significantly larger. We find that this difference in adopter size basically accounts for the two studies' differing sample-wide ROA results. Our analyses indicate that in both studies' samples, JIT adopters below a minimum firm-size threshold achieve no relative ROA improvement, whereas firms above this threshold (i.e., medium-size to large firms) do achieve relative improvement. These findings support both Alles et al.'s (1995) contention that JIT's benefits may be greatest where there are substantial asymmetries between workers and management, as well as BLV's observation that firms with more market power may retain more of the financial benefits of JIT adoption.

The remainder of this paper is organized as follows. In the next section, we develop four hypotheses regarding associations between JIT adoption and changes in financial performance. Section III describes our sample selection procedures and provides sample descriptive statistics. Section IV explains our hypotheses tests and reports test results. Section V describes sensitivity tests and related results, and reconciles our findings with BLV's. The final section summarizes the paper and offers suggestions for future research.

II. HYPOTHESES

Profitability Hypotheses

McIlhattan (1987, 23) defines JIT as "the constant and relentless pursuit of the elimination of waste, with waste being defined as anything that does not add value to a product--inspection, queue time, and stock." Lubben (1988, 13) asserts that "the greatest misconception about JIT is that it is an inventory control system. Although structuring a system for JIT will control inventory, that is not its major function." Thus, our primary tests employ a comprehensive measure of firm performance, ROA, instead of a measure capturing only inventory management. ROA (income / average total assets) is the product of profit margin (income / sales) and asset turnover (sales / average total assets). Although interdependent, profit margin and asset turnover reflect different determinants of a firm's success or failure. Atkinson et al. (2001, 543) describe asset turnover as a measure of productivity--the ability to generate sales with a given level of investment, and profit margin as a measure of efficiency--the ability to control costs at a given level of sales activity.

We expect JIT adoption to be associated with improved asset turnover. A JIT adopter's ability to turn its assets should increase with improved product quality, greater responsiveness to customer demand because of shorter lead times, and the greater product line variety that is sustainable with increased manufacturing flexibility. In addition, lower inventory also reduces the denominator used to calculate total asset turnover. (2)

We also expect that JIT is associated with increased profit margins. JIT utilization reveals activities that add no value. These activities and their related costs are often either hidden by large inventory stocks, or are ignored because buffer stocks are a convenient "solution" to production line or other system failures. Without excess inventory to mitigate these problems, JIT adopters are more likely to develop cost-saving solutions, thereby increasing profit margin. Alles et al. (1995, 188) note:

   Eliminating buffer inventories makes the production setting "transparent,"
   exposing flaws and thus helping management and workers to eliminate
   problems. The incentive to eliminate these problems is greater when
   inventories are low, because small buffer stocks provide less insurance
   against problems. Management's role is to support and coach workers. By
   providing this support, managers get better information about problems
   faced by workers and the ways workers overcome these problems.

The preceding discussion suggests that JIT adopters increase ROA and both of its components, relative to the performance of non-adopting control firms:

H1: Pre- to post-adoption change in ROA is greater for JIT adopters than for matched non-adopters.

H2: Pre- to post-adoption change in asset turnover is greater for JIT adopters than for matched non-adopters.

H3: Pre- to post-adoption change in profit margin is greater for JIT adopters than for matched non-adopters.

Predominant Source of JIT-Related ROA Performance

We extend prior literature by considering whether relative changes in JIT adopters' ROAs result primarily from relative changes in efficiency (i.e., profit margin) or from relative changes in productivity (i.e., asset turnover). Because we cannot predict which effect dominates, our fourth hypothesis is nondirectional:

H4: Relative changes in profit margin and asset turnover are equally significant in explaining relative ROA changes associated with JIT adoption.

III. SAMPLE SELECTION AND DESCRIPTIVE STATISTICS

Sample Selection

Table 1 summarizes our sample selection procedures, which included searches of both public data sources and existing literature. Inclusion in the initial adopter sample required only that "JIT" or some variation thereof appear in a publicly available document (e.g., an annual report or news story) related to a given firm. The initial search identified 623 firms, of which 445 had at least five years of sales data available on Compustat. After carefully reading the 445 source documents that tentatively identified firms as JIT adopters, we concluded that 130 firms had not integrated JIT into their operations; we deleted these firms from the sample. Of the remaining 315 firms, we deleted 13 because we could not determine an inventory valuation method (4 firms) or because we could not match (as discussed below) on inventory valuation method at the two-digit level or higher (9 firms). Our preliminary sample, therefore, includes 302 JIT adopters.

We matched non-adopting firms to the 302 JIT adopters in the preliminary sample. We matched first on inventory valuation method, and then attempted to match on net sales in the year preceding JIT adoption at the four-digit industry level. If an acceptable match was not available, then we dropped to the three-digit level and, if necessary, to the two-digit level. (3) We followed BLV in defining pre-adoption and post-adoption windows and deleted 101 firm-pairs for which either the JIT adopter or its matched control firm lacked Compustat data to calculate inventory turnover and ROA for each year from -3 to +3, relative to the JIT adoption year. Thus, the final sample includes 201 JIT adopters and matched counterparts.

Three features of our sample selection may affect our test results. First, firms often implement JIT incrementally, and the stage and scope of implementation can be difficult to determine from public sources. Thus, for some firms, our "JIT adoption year" is an estimate based on our reading of adopters' source documents. Second, our tests, like those of prior studies, may be subject to selection bias. If the firm characteristics that led to JIT adoption also led to superior future performance for reasons unrelated to JIT, then any performance effects associated with JIT adoption may actually be caused by such characteristics. Although sensitivity analyses reported in Section V yield no evidence that our inferences are affected by selection bias, limitations in the research design preclude us from completely ruling out such bias as an alternative explanation for our results. Third, our sample could also suffer from a disclosure bias if only successful adopters disclose implementation of JIT in their annual reports. Our sample of adopters excludes any firms that adopted JIT, quickly deemed it a failure, and elected not to discuss the failure in their adoption-year annual reports. While this possibility creates a bias toward JIT-associated profitability in our sample, it is unlikely that the number of unsuccessful JIT adopters reaching such immediate conclusions would be large enough to account for our study's results.

Descriptive Statistics

Table 2, Panel A, provides the JIT adoption year distribution for the sample of 201 JIT adopters. The weighted-average adoption year is 1988. Table 2, Panel B reports the two-digit industry distribution for the adopter and control samples. Forty-three percent of all firm-pairs are in industries 35 or 36 (Industrial Equipment and Commercial Machinery [including computers] and Electronic Equipment, respectively). As reported in Table 2, note a, 56.7 percent of the 201 firm-pairs use LIFO to value at least some portion of total inventory.

We align the annual data of each JIT adopter and its matched non-adopter in event time based on the adoption year (year 0). Like BLV, we define the pre-adoption (post-adoption) period as the three-year period preceding (following) adoption. Following BLV, we dampen the influence of extreme observations by winsorizing at the 5th and 95th percentiles of a variable's pooled distribution (six years of pre- and post-adoption data for 402 firms). (4) For each firm, we then measure a pre- (post-) adoption variable as its median value in the three-year pre- (post-) adoption period.

Table 2, Panel C provides means and medians of ten pre-adoption attributes for the JIT and control samples. The last four columns of Panel C are mean and median paired differences between subsamples, standard deviations of paired differences, and significance levels from Wilcoxon signed rank tests. (5) Paired differences in total assets, net sales, and total inventory indicate that JIT adopters are, on average, larger than their non-adopting counterparts. Aside from this size difference, the only other significant difference in Panel C is for fixed cost (depreciation / cost of sales) ratios, with JIT adopters having the larger ratios.

Sample Validation

To validate our sample selection procedures, we examine whether our JIT sample firms differ from non-adopting matched firms in ways consistent with JIT adoption. In Figure 1, we plot median total inventory turnover (cost of sales / average total inventory) for both samples in years -3 through +3. The graph indicates that adopters increased total inventory turnover (both absolutely and in comparison to non-adopters), and also suggests that adoption years are reasonably well identified. Although inventory turnover mildly increases in year -1, the first substantial increase occurs in year 0, followed by sustained increases in subsequent years.

To validate our sample selection procedures more formally, we conduct Wilcoxon signed rank tests of the paired differences in changes in total inventory turnover and investment:

(1) [DIF[DELTA]INVTURN.sub.J] = [[DELTA]INVTURN.sub.J] - [[DELTA]INVTURN.sub.C]

(2) [DIF[DELTA]INVASSET.sub.J] = [[DELTA]INVASSET.sub.J] - [[DELTA]INVASSET.sub.C]

where:

J = JIT adopter;

C = matched control firm;

[DELTA]INVTURN = post-adoption total inventory turnover minus pre-adoption total inventory turnover; and

[DELTA]INVASSET = post-adoption total inventory / total assets minus pre-adoption total inventory / total assets.

Table 3 results confirm that JIT adopters achieve greater increases in total inventory turnover and greater reductions in inventory investment. For the JIT sample, both changes are significant at p [less than or equal to] 0.001 in one-tailed signed rank tests. Tests of paired differences between the JIT and control samples indicate that pre-adoption differences in inventory turnover and investment are not significant. However, in the post-adoption period, JIT adopters have higher inventory turnover and lower inventory investment (p = 0.012 and 0.002, respectively). Finally, tests of paired differences in changes indicate that JIT adopters increased inventory turnover and reduced inventory investment, compared to their matched counterparts (p [less than or equal to] 0.001 in both cases). (6)

IV. H1-H4 TESTS AND TEST RESULTS

Tests of the ROA, Asset Turnover, and Profit Margin Hypotheses (H1-H3)

Our tests of changes in ROA, asset turnover, and profit margin are analogous to the sample validation test:

(3) [DIF[DELTA]ROA.sub.J] = [DELTA]ROA - [[DELTA]ROA.sub.C]

(4) [DIF[DELTA]ASSETTURN.sub.J] = [[DELTA]ASSETTURN.sub.J] - [[DELTA]ASSETTURN.sub.C]

(5) [DIF[DELTA]MARGIN.sub.J] = [[DELTA]MARGIN.sub.J] - [[DELTA]MARGIN.sub.C]

where:

J and C = JIT and control firms, respectively;

[DELTA]ROA = post-adoption ROA minus pre-adoption ROA, where ROA is income before extraordinary and special items / average total assets; (7)

[DELTA]ASSETTURN = post-adoption asset turnover minus pre-adoption asset turnover, where asset turnover is net sales / average total assets; and

[DELTA]MARGIN = post-adoption profit margin minus pre-adoption profit margin, where profit margin is income before extraordinary and special items / net sales.

We test H1-H3 with one-tailed signed rank tests of DIF[DELTA]ROA, DIF[DELTA]ASSETTURN, and DIF[DELTA]MARGIN, respectively.

ROA (H1) Results

Table 4 indicates that JIT adopters' ROAs declined an average of 3.6 percent ([0.054 - 0.056] / 0.056) from the pre- to post-adoption period, which is significant in a two-tailed signed rank test (p = 0.055). However, non-adopters' ROAs decline much more--by 21.1 percent (p = 0.001). (8) The last two columns of Table 4 report tests of paired differences. In the pre-adoption period, there was no significant difference between adopters and control firms (p = 0.861), but adopters' ROAs were higher than control firms' ROAs in the post-adoption period (p = 0.010). Most importantly, the results support H1; JIT adopters' ROAs increase relative to control firms' ROAs (p = 0.010). (9)

We also tested paired ROA changes separately for years 0 through +3. For these years, median paired differences in ROA changes were 0.006, 0.007, 0.007, and 0.003, respectively. Thus, JIT adopters' relative ROA improvement deteriorates by the end of the post-adoption period. We conducted additional tests on years +4, +5, and +6 using 128 firm-pairs for which data were available through year +6. For year +4, the paired differences in ROA changes were significant at the 0.078 level; for years +5 and +6, the paired differences were insignificant. This apparent dissipation of the association between JIT and financial performance is consistent with BLV's (1996, 187) argument that, in a competitive market, firms will mimic early JIT adopters if JIT adoption improves performance. In the long run, any gains realized by early adopters will be competed away. Thus, control firms may have ultimately adopted JIT, but made no related disclosures because JIT adoption was no longer newsworthy at the time of their late adoptions. (10) Consistent with this scenario, paired differences in changes in inventory turnover are also no longer significant by year +6, although inventory turnover for both samples improved greatly.

Asset Turnover (H2) and Profit Margin (H3) Results

Results from tests of H2 and H3 appear in Table 5. JIT adopters' pre- to post-adoption asset turnover changes were statistically insignificant, whereas asset turnover declined significantly for non-adopters (p = 0.013). The tests of pre- to post-adoption changes in profit margin indicate that declines occurred in both samples (p = 0.114 and 0.001 for adopters and non-adopters, respectively). Results of paired tests indicate there were no pre-adoption differences in asset turnover and profit margin between samples. Post-adoption results reveal that profit margin is significantly higher for adopters (p = 0.010). Most important, our H2 prediction that relative asset turnover would improve for JIT adopters is supported (p = 0.033), and our H3 prediction that JIT adopters would enjoy a higher relative change in profit margin is also supported (p = 0.014). (11)

Tests of the Dominant Source of Post-Adoption ROA Performance (H4)

Tests of H4 examine the relative effects of profit margin and asset turnover changes on ROA changes. We first construct two variables that capture the ROA effect of relative changes in profit margin and asset turnover:

(6) [MARGIN_EFFECT.sub.J] = [DIF[DELTA]MARGIN.sub.J] x [PRE_ASSETTURN.sub.J]

(7) [TURNOVER_EFFECT.sub.J] = [DIF[DELTA]ASSETTURN.sub.J] x [PRE_MARGIN.sub.J]

where DIF[DELTA]MARGIN and DIF[DELTA]ASSETTURN are as previously defined, PRE_ASSETTURN is adopter j's pre-adoption asset turnover, and PRE_MARGIN is adopter j's pre-adoption profit margin.

With a signed rank test of MARGIN_EFFECT, we examine the ROA effect of relative profit margin changes, when JIT adopters' asset turnovers are held constant at pre-adoption levels. We conduct a similar test for asset turnover. Our test of H4 examines within-adopter differences in MARGIN_EFFECT and TURNOVER_EFFECT:

(8) [DIF_EFFECT.sub.j] = [MARGIN_EFFECT.sub.j] - [TURNOVER_EFFECT.sub.j].

Significantly positive DIF_EFFECT would suggest that relative profit margin changes, compared to relative asset turnover changes, more positively influence JIT adopters' post-adoption relative ROA changes. Significantly negative DIF_EFFECT would suggest the converse is true. We test DIF_EFFECT with a two-tailed signed rank test.

Source of ROA Performance (H4) Results

Table 6 reports results of H4 tests conducted on full (n = 201), once-reduced (n = 176), and twice-reduced (n = 162) samples. The once-reduced sample excludes the 25 JIT adopters reporting pre-adoption losses. For these 25 firms, TURNOVER_EFFECT is negative for adopters with relative increases in asset turnover, and positive for adopters with relative decreases in asset turnover. Although these effects are consistent with the DuPont formula (e.g., an unprofitable firm that increases its asset turnover will decrease ROA), the presence of these firms in the full sample may understate the effect of asset turnover on JIT adopters' relative ROA changes. In addition, excluding these 25 firms eliminates the effect of mean reversion on the part of those JIT adopters that were least successful during the pre-adoption period (although, arguably, such firms also have the most to gain from JIT-induced improvement in cost control). The twice-reduced sample excludes these 25 firms (thus purging the somewhat perverse TURNOVER_EFFECT) and the 14 JIT adopters paired with non-adopters reporting pre-adoption losses (thereby applying any concern regarding mean reversion uniformly to the adopter and control samples).

In the tests of the full sample, median MARGIN_EFFECT is significantly positive (p = 0.014), but median TURNOVER_EFFECT is insignificant (p = 0.322). Most important, median DIF_EFFECT is positive (0.007) and significant at the 0.032 level in a two-tailed test, indicating that profit margin changes, compared to asset turnover changes, contribute more to JIT adopters' relative ROA improvement. As might be expected, the tests conducted on the once-reduced sample do not fully support the primacy of profit margin in JIT adopters' relative ROA improvement--median DIF_EFFECT, though positive, is not significant (p = 0.259). Tests of the twice-reduced sample suggest that relative improvement in profit margin and asset turnover both contribute to relative ROA improvement (p = 0.015 and 0.063, respectively), but that the ROA effect of profit margin changes exceeds that of asset turnover changes (p = 0.052 in the two-tailed test). Overall, the results in Table 6 suggest that the positive association between JIT and ROA performance results primarily from JIT adopters' relatively improved profit margins, consistent with Alles et al.'s (1995) contention that JIT promotes transparent operations and cost-saving improvements in production processes.

V. SENSITIVITY TESTS AND RECONCILIATION WITH BLV

Self-Selection Sensitivity Test

Our primary tests ignore self-selection inherent in the JIT adoption decision. Therefore, if the underlying firm characteristics that prompt JIT adoption are also associated with superior future performance (without regard to JIT's performance effects), then our tests of paired differences in changes in ROA, asset turnover, and profit margin may overstate JIT's performance effects. To evaluate the effects of endogeneity on our H1 (ROA) inferences, we conduct two-stage self-selection analysis similar to that described by Maddala (1977, 1983, 1991) and used by Hogan (1997) in a study of auditor selection. The first stage in the analysis requires estimation of a probit model of JIT choice. (12) We estimated the following model:

(9) [ADOPT.sub.j] = [[beta].sub.0] + [[beta].sub.1][PRE_SALES.sub.j] + [[beta].sub.2][PRE_INVTURN.sub.j] + [[beta].sub.3][VAR_DEMAND.sub.j] + [[beta].sub.4][INNOVATE.sub.j] + [u.sub.j]

where:

ADOPT = 1 if firm j is a JIT adopter, and 0 otherwise;

PRE_SALES = median sales in the pre-adoption period;

PRE_INVTURN = median total inventory turnover in the pre-adoption period;

VAR_DEMAND = variability of demand for the firm's output; and

INNOVATE = firm innovativeness, proxied by median R&D / Sales in the pre-adoption period.

We include PRE_SALES because average firm size differs between samples, and because firms with greater resources may be more inclined to adopt new technologies. We include PRE_INVTURN because managers of firms with low inventory turnover may anticipate the greatest benefits from JIT adoption, as BLV suggest. We include VAR_DEMAND because JIT may be less beneficial to firms for which product demand (and thus production and input demand) is more variable. To estimate VAR_DEMAND, we regress a firm's cost of sales for the 24 quarters preceding JIT adoption on a quarterly time-trend variable. (13) VAR_DEMAND is the standard deviation of the regression residuals, scaled by mean cost of sales over the estimation period. We include INNOVATE to reflect the possibility that innovative firms, proxied by R&D spending scaled by sales, may be more likely to adopt new technologies. Finally, we estimate the model with samples matched on industry, inventory method, and size.

The upper portion of Table 7 reports results from estimating the first-stage model. Coefficient estimates all have the expected signs, and the coefficients of PRE_SALES and INNOVATE are significant (p = 0.050 and 0.010, respectively). In addition, the model is significant at the 0.001 level.

In the second stage of the selectivity analysis, we estimated, separately for the two samples, the following model (firm subscripts are suppressed):

(10) [DELTA]ROA = [[alpha].sub.0] + [[alpha].sub.1]PRE_SALES + [[alpha].sub.2]PRE_INVTURN + [[alpha].sub.3]VAR_DEMAND + [[alpha].sub.4]INNOVATE + [[alpha].sub.5][M.sub.J(C)] + [epsilon]

where [DELTA]ROA, PRE_SALES, PRE_INVTURN, VAR_DEMAND and INNOVATE are as previously defined, and [M.sub.J] and [M.sub.C] are selectivity variables. In the adopter equation, [M.sub.J] is -f([[beta]'Z) / F([[beta]'Z); for non-adopters, [M.sub.C] is f([[beta]'Z) / (1 - F([[beta]'Z)), where f(*) and F(*) are the density and distribution functions, respectively, of the standard normal distribution, and [beta]'Z is the prediction from the first-stage probit model.

We assess the role of selection bias by examining the estimates of [[alpha].sub.5] in the two equations. If [[alpha].sub.5] < 0 in both equations, then our primary results overstate JIT's ROA effect because [DELTA]ROA is overstated for the adopter sample and understated for the control sample. However, the coefficients of [M.sub.J] and [M.sub.C], reported in the lower portion of Table 7, are negative and positive, respectively, and not significant at conventional levels. Thus, the analysis provides no evidence that self-selection bias affects our inferences. Likewise, we found no evidence of such bias in analogous H2 and H3 analyses using [DELTA]ASSETTURN and [DELTA]MARGIN as dependent variables in the second stage. (14)

Reconciliation with BLV

In contrast to BLV's findings, the results in Table 4 suggest that JIT adoption is associated with superior ROA performance, and sensitivity analyses yield no evidence that this finding is attributable to between-sample differences in firm size, cost structure, or LIFO liquidation income (see footnote 14), or to self-selection bias. In an effort to understand the between-study difference in sample-wide results, we conducted the following analyses.

Differences in Research Methods and Statistical Power

Our method, analysis of paired differences, differs from BLV's ANOVA. To examine the effect of this difference, we first ran our test on BLV's sample. Like BLV, we found no relative ROA improvement for JIT adopters. Next, we used BLV's method on our sample and observed a significant relative increase in ROA for JIT adopters. Thus, the difference in results is not method-driven, but instead is attributable to either (1) the greater statistical power of our test (i.e., our much larger sample), or (2) a fundamental difference in samples between the studies.

We ran an informal bootstrap analysis on our sample to evaluate the effect of the between-study difference in statistical power. We drew 100 random samples of size 46 (BLV's sample size) from our sample of 201 firm-pairs, and conducted our test of DIF[DELTA]ROA on each sample. The median significance level in these tests was 0.105 (BLV's significance level in their analogous ROA test was 0.26), and 30 of 100 samples had p-values less than 0.05. Because we continue to find a reasonably significant association between JIT and ROA using these 46 firm-pair subsets of our full sample, we conclude that fundamental differences between the two studies' samples, rather than statistical power, primarily account for the studies' differential inferences.

Tests Examining Customer Concentration and Span-of-Operations Effects

BLV report significant relative ROA improvement among "free" JIT adopters not disclosing a major customer under SFAS No. 14, but no relative improvement among "captive" adopters disclosing the existence of a major customer. To enhance their customer concentration tests, BLV required that adopter and control firms match on customer concentration, and that firms in their samples report operations spanning fewer than four three-digit SICs. Ex ante, we elected not to impose these restrictions, thereby avoiding any resulting loss of generalizability and reduction in sample size. However, our sample includes enough firm-pairs meeting the BLV restrictions to determine whether customer concentration or span-of-operations effects likely account for differences between our results and BLV's.

We conducted an ROA test on 114 firm-pairs that match on customer concentration, 23 captive and 91 free, and on a "quasi-matched" sample that includes these 114 firm-pairs plus 35 additional firm-pairs in which captive adopters are matched with free non-adopters. Under BLV's hypothesis, we expect at least as much downstream sharing of JIT-induced cost savings in the latter 35 firm-pairs as in captive adopter/captive non-adopter firm-pairs. Thus, the test conducted on the quasi-matched sample is biased toward finding BLV's result. We regress DIF[DELTA]ROA on CAPTIVE (coded 1 if the JIT adopter in a firm-pair is captive, 0 otherwise), and two control variables: paired differences in pre-adoption sales (DIF_PRE_SALES), and paired differences in pre-adoption fixed-cost ratios (DIF_PRE_FCRATIO). BLV concluded that captive JIT firms did not enjoy relative improvement in ROA, whereas free adopters did. If this result holds in our sample, the regression equation's intercept will be significantly positive, and CAPTIVE will have a significantly negative coefficient.

The results reported in Table 8 reveal that CAPTIVE is insignificant in both regressions, suggesting that customer concentration plays no role in JIT adopters' relative ROA changes. Moreover, the significantly positive intercepts in the equations, along with the (insignificantly) positive coefficients on CAPTIVE, indicate that paired differences in ROA changes are significant for both free and captive JIT adopters, after controlling for between-sample differences in firm size and cost structure. (15)

To analyze the sensitivity of our results to a span-of-operations screen, we tested DIF[DELTA]ROA using only those firm-pairs in which both firms operated in fewer than N three-digit SIC classes, where N, in sequential tests, was set equal to 3, 4, 5, and 6. With N = 3 (N = 4), the sample size was 60 (89) firm-pairs, and the p-value of the ROA test was 0.049 (0.108); with N = 5 (107 firm-pairs), the p-value was 0.037; with N = 6 (124 firm-pairs), the p-value was 0.026. Thus, our ROA results are not sensitive to BLV's span-of-operations screen. (16)

Tests Examining Adopter Size Effects

A careful comparison of JIT adopter profiles in the BLV and present studies reveals two primary differences. First, our sample firms are much larger, on average, than BLV's. Median pre-adoption sales for our adopters are $397.5 million, but only $108.5 million for BLV's adopters. (17) Second, Table 2, Panel C indicates that our adopters have higher fixed-cost ratios than our non-adopters (p = 0.002). In contrast, our analysis of the BLV samples indicates that BLV non-adopters have the higher fixed-cost ratios (p = 0.047). Thus, if JIT's benefits are greater for large firms and/or firms with large fixed-cost ratios, then our results may differ from BLV's because our adopter sample contains more large firms and firms with high fixed-cost ratios than does BLV's sample. (18)

To examine the effect of adopter size on our H1 (ROA) results, we first partitioned our sample and the BLV sample into large-adopter and small-adopter samples using the median pre-adoption sales in BLV as the partition. We find no relative ROA improvement in either of the two small-firm samples (N = 23 and 39 in BLV's and our sample, respectively). The median paired difference in change in ROA is -0.010 (-0.011) in BLV's (our) small-adopter sample, and is not significant (p = 0.477 and 0.462, respectively). However, in the large-firm samples (N = 23 and 162 in BLV's and our sample, respectively), the median paired differences are 0.021 in the BLV sample and 0.014 in our sample (p = 0.031 and 0.005, respectively).

Next, we considered whether adopter-size is an important intervening variable in the sample-wide ROA results reported in Table 4. We partitioned our adopter sample on its own (rather than BLV's) median pre-adoption sales value ($397.5 million) and re-ran our test on the resulting small- and large-adopter samples. The results indicated that both small and large JIT adopters enjoy ROA performance superior to that of non-adopters (p = 0.052 and 0.045, respectively). Further, the correlation between relative ROA improvement and adopter size for our full sample of 201 adopters is insignificant (p = 0.322), and multiple regression models yield no evidence of an interaction between JIT adoption and firm size in explaining our sample firms' ROA changes.

Thus, although an adopter-size effect apparently reconciles our results to BLV's, the absence of a significant size effect across our sample of 201 adopters suggests that the effect is not linear, but is instead limited to firms below a firm-size threshold. As noted previously, BLV's 23 largest adopters enjoy significant relative ROA improvement, on average, whereas their 23 smallest adopters do not. Among BLV's 23 smallest adopters, median pre-adoption sales are $33 million; among their 23 largest adopters, median pre-adoption sales are $322 million. Hence, comparing the two groups' relative ROA changes is clearly a comparison of very small firms to other firms. When our sample is partitioned on BLV's median sales value, our 39 very small adopters (median sales of $53 million) do not enjoy significant relative ROA improvement, while the 162 other adopters (median sales $581 million) do, and the inference regarding adopter size and relative ROA changes is identical. Partitioning our sample on its own median sales value again results in two samples with strikingly different median sales values ($127 million vs. $1,313 million). Yet, as reported above, both groups enjoy relative ROA gains, evidently because, under this partitioning, the small-adopter group is less dominated by very small firms. Collectively, these results suggest that the difference in inferences between the two studies arises primarily from the much greater presence of very small firms in BLV's sample. (19)

Two explanations for the apparent ineffectiveness of JIT for very small adopters seem plausible. First, Alles, et al. (1995, 196) note that their results "suggest that JIT is more likely to be useful for purposes other than inventory management in circumstances where there are substantial information asymmetries between workers and management." If such information asymmetry is lacking in very small firms, then very small JIT adopters, which constitute a greater proportion of the BLV sample than our sample, are unlikely to realize significant benefits, apart from improved inventory utilization. Second, the adopter-size result is consistent with BLV's suggestion that firms with limited market power may be less able to retain any financial benefits of adopting JIT.

Tests Examining Adoption Timing Effects

We considered whether our inclusion of 78 firms that adopted JIT after 1989 (BLV's latest adoption year) contributes to the two studies' differing ROA results. We partitioned our sample into 123 early (pre-1990) and 78 late (post-1989) adopters and re-ran our ROA test. For early adopters, paired differences in ROA changes were significant (p = 0.007), but relative ROA changes were not significant for late adopters (p = 0.218). Thus, these test results do not reconcile the two studies' results, but instead are more consistent with a "first mover" advantage, and with a scenario in which firms with the most to gain from JIT are among the earliest to adopt.

Adopter Size and Adoption Timing Effects in H2-H4

Finally, we examined the effects of adopter size and adoption timing on the association between JIT adoption and relative changes in asset turnover and profit margin. With the sample partitioned on BLV's median pre-adoption sales, the results mirror those described for ROA. DIF[DELTA]ASSETTURN, DIF[DELTA]MARGIN, and DIF_EFFECT are all reasonably significant in the large-firm sample (p [less than or equal to] 0.067), but insignificant in the small-firm sample (p [greater than or equal to] 0.164). With our sample partitioned on its own median pre-adoption sales, DIF[DELTA]MARGIN is reasonably significant in both samples (p [less than or equal to] 0.097), but DIF[DELTA]ASSETTURN is significant in the small-adopter sample (p = 0.049) and insignificant in the large-adopter sample (p = 0.214). Consistent with these results, we also find that DIF_EFFECT is significant for the large-adopter sample (p = 0.043), but is insignificant for the small-adopter sample (p = 0.217).

In adoption timing tests using ROA components, we find somewhat divergent results for DIF[DELTA]ASSETTURN and DIF[DELTA]MARGIN. DIF[DELTA]ASSETTURN is insignificant for early adopters (p = 0.170), but significant for late adopters (p = 0.034). Conversely, DIF[DELTA]MARGIN is significant for early adopters (p = 0.012), but insignificant for late adopters (p = 0.230), and DIF_EFFECT is also significant for early adopters (p = 0.037) and insignificant for late adopters (p = 0.386). (20)

VI. CONCLUSION

In this study, we use a large sample of JIT adopters and matched non-adopters to examine the association between JIT adoption and financial performance. Our study makes several contributions to the literature.

First, we find that JIT adopters outperformed matched firms in ROA improvement over a three-year post-adoption period. JIT adopters' pre- to post-adoption ROA changes exceeded those of matched non-adopters by an average of 1.0 percent--a highly significant difference in our primary test. However, the results of additional analyses suggest that very small firms may realize minimal benefits from JIT, a finding that reconciles our inferences with those of BLV, whose sample included a much greater proportion of such firms. The finding that very small JIT adopters achieve no significant relative ROA improvement is consistent with Alles et al.'s (1995) observation that JIT most benefits firms with significant information asymmetry between workers and managers, and with BLV's prediction that a firm's market power may play a significant role in determining its ability to retain JIT-related financial benefits.

The evidence of a sample-wide association between JIT adoption and ROA improvement is from tests incorporating three-year pre- and post-adoption windows. However, the results of additional analyses suggest that relative ROA improvement is concentrated among the earliest JIT adopters (which may have the most to gain from JIT adoption), and that by the fifth or sixth year following JIT adoption, adopters no longer exhibit superior performance in inventory turnover or ROA. Taken together, these results suggest a "first mover" advantage for early adopters, followed by dissipation of the advantage as JIT becomes more widely adopted.

BLV reported post-adoption relative ROA increases only among those JIT adopters not reporting the existence of a major customer under SFAS No. 14. BLV infer that relatively powerful customers expropriate JIT's benefits from adopters with concentrated customer bases. In contrast, we find no evidence that the realization and retention of JIT-related benefits depends on a diverse customer base.

BLV contend that powerful customers may capture adopters' JIT-related cost savings. However, JIT requires substantial coordination with upstream suppliers and downstream customers in ordering, production scheduling, and delivery. If downstream coordination costs are increasing in the number of customers with whom JIT adopters must maintain relationships, then "captive" JIT adopters may actually enjoy lower downstream coordination costs. Thus, an SFAS No. 14 disclosure may proxy for two constructs (a firm's relative market power and its coordination costs) expected to have opposite effects on post-adoption earnings. Future research may test the customer concentration and coordination cost hypotheses, particularly if more suitable proxies for the two constructs can be identified.

Our sample-wide results suggest that JIT adopters, compared to non-adopters, improve both components of ROA--profit margin and asset turnover. However, we extend prior literature by identifying the primary source of JIT adopters' relative ROA increases. Our sample-wide evidence that improved profit margins are the main source of post-JIT adoption ROA improvement (1) is consistent with Alles et al.'s (1995, 188) contention that JIT makes the production setting "transparent," assisting line workers and management alike in making cost-saving improvements, and (2) indicates that JIT's benefits are not limited to reduced inventory holding costs and reduced total investment arising from leaner inventories. (21)

Our study has several limitations. Our self-selection bias analysis relies on an untested model of JIT choice. Thus, we cannot rule out the possibility that our reported results stem at least partially from underlying dissimilarities between JIT adopters and non-adopters. We leave to future research more sophisticated modeling of JIT choice and further study of the potential endogeneity in the JIT adoption decision. The interdependence of profit margin and asset turnover increases the difficulty of assessing the two measures' relative importance in the association between JIT adoption and ROA performance. Thus, our analyses regarding the source of the positive association between JIT adoption and ROA performance are only an initial attempt to refine our understanding of JIT's performance effects. Finally, our study covers a period in which American companies adopted a number of initiatives aside from JIT, such as cellular manufacturing, activity-based costing, and total quality management. Concurrent adoptions of JIT and other initiatives require that readers exercise caution when interpreting our study's results.

TABLE 1
Sample Selection and Screening

                                                       Sample Size

Searched Lexis/Nexis COMPANY file (a)                      408
Searched Lexis/Nexis ALLNEWS file (a)                      142
Searched NAARS (a)                                           9
Additional firms identified during the literature           55
  review
Additional firms provided by Balakrishnan et al.             9
  (1996)
  Total potential sample firms                             623
Firms without five years of data available on             -178
  Compustat (1977-1995) (b)
  Firms for which we examined source documents             445
   (e.g., annual reports)
Firms eliminated because source indicated only a          -130
  passing reference to JIT or a JIT adoption date
  outside the range of years for which Compustat
  data are available (c)
Firms for which inventory valuation method was              -4
  not available on Compustat
Firms for which we could not find a match on                -9
  Compustat (d)
  Preliminary sample                                       302
Firms missing data necessary to calculate total           -101
  inventory turnover and ROA in each year from -3
  through +3 relative to the JIT adoption year
  Final sample                                             201

(a) The COMPANY file included detailed 10-Ks back to 1. We searched
the NAARS and ALLNEWS files to obtain pre-1987 adopters (although
NAARS does not contain the Management Discussion & Analysis portion
of firms' annual reports). The search string was ((just in time) or
(JIT) or (pull system) or (continuous flow manufacturing) or (zero
inventor!)) w/10 ((implement!) (chang!) or (switch!) or (adopt!) or
(enhanc!) or (expand!) or (extend!)).

(b) Many firms identified in the ALLNEWS file and during the
literature review were  private firms or subsidiaries or divisions
of other firms. 151 of the 178 firms that this screen eliminated
had no Compustat data.

(c) Passing references to JIT ranged from use of the phrase "just
in time" in contexts unrelated to JIT, to discussions regarding
supplying customers on a JIT basis with no indication that the firm
had adopted JIT practices in its own operations. We also eliminated
two firms that appeared to be unlikely candidates for implementing
JIT. One was in SIC 4833 (Television Broadcast Stations); the other
was in SIC 9995 (Nonoperating Establishments).

(d) These firms could not be matched on inventory valuation method
at the two-digit industry level.
TABLE 2
Descriptive Statistics for 201 JIT Adopters and Matched Control
Firms (a)

Panel A: Distribution of JIT Adoption Years

Year    Number of firms      %

1982            3            1.5
1983            6            3.0
1984           12            6.0
1985           15            7.5
1986           17            8.5
1987           18            9.0
1988           27           13.4
1989           25           12.4
1990           26           12.9
1991           22           10.9
1992           16            8.0
1993           14            7.0

              201          100.0

Panel B: Distribution of Two-Digit Industry Classifications

                                                  Number of
Two-Digit Industry Code   Industry Description      Firms        %

           20             Food                          1       0.5
           22             Textile mill products         2       1.0
           23             Apparel                       1       0.5
           24             Lumber                        1       0.5
           25             Furniture                     8       4.0
           26             Paper                         2       1.0
           27             Printing, publishing          4       2.0
           28             Chemicals                     5       2.5
           30             Rubber and plastics           5       2.5
           31             Leather                       2       1.0
           32             Glass, pottery                1       0.5
           33             Primary metals               12       6.0
           34             Fabricated metals            11       5.5
           35             Industrial equipment         46      22.9
           36             Electronic equipment         40      19.9
           37             Motor vehicles               16       8.0
           38             Instrumentation              23      11.4
           39             Other manufacturing           6       3.0
           50             Wholesale durables            4       2.0
           53             Department stores             3       1.5
           54             Grocery stores                2       1.0
           56             Clothing stores               1       0.5
           57             Furniture stores              1       0.5
           59             Miscellaneous retail          2       1.0
           73             Packaged software             2       1.0

                                                      201     100.0

Panel C: Financial Attributes of the JIT Adopter and Control
Samples

                          JIT Sample         Control Sample

Firm Attribute (b)      Mean     Median      Mean     Median

Total assets          1,736.1     303.3      988.0     169.0
Net sales             2,365.0     397.5    1,185.8     234.3
Total inventory         348.4      71.2      196.9      43.0
Gross margin            0.349     0.332      0.333     0.312
Operating ROA           0.107     0.110      0.107     0.104
ROA                     0.056     0.056      0.057     0.057
Inventory / assets      0.251     0.236      0.257     0.249
Inventory turnover      4.041     3.437      4.200     3.601
Debt / assets           0.486     0.488      0.492     0.493
Fixed-cost ratio        0.062     0.053      0.052     0.044

                                  Paired Difference

                                             Std.         p-
Firm Attribute (b)      Mean      Median     Dev.      value (c)

Total assets            748.1       66.2    3,876.5      0.001
Net sales             1,179.2       79.0    5,803.9      0.001
Total inventory         151.6       16.4      679.5      0.001
Gross margin            0.016      0.014      0.136      0.108
Operating ROA           0.000      0.004      0.086      0.978
ROA                    -0.001      0.001      0.063      0.861
Inventory / assets     -0.006      0.000      0.120      0.472
Inventory turnover     -0.159      0.000      2.398      0.592
Debt / assets          -0.006     -0.029      0.221      0.504
Fixed-cost ratio        0.009      0.004      0.038      0.002

(a) The adopter and control samples match on inventory valuation
method. We coded firms as LIFO users if any digit in Compustat item
#59 indicated LIFO usage; otherwise, we coded firms as FIFO users.
114 (56.7 percent) of all firm-pairs are LIFO users.

(b) Firm attribute measures are median values in the three-year
period preceding the adoption year. Total assets, net sales, and
total inventory are stated in $ millions. Operating ROA is
calculated with operating income after depreciation in the
numerator, whereas ROA is calculated with income before
extraordinary and special items in the numerator (as in BLV). Both
measures use average total assets in the denominator. Inventory /
assets = total inventory / total assets. Inventory turnover = cost
of sales / average total inventory. Debt / assets = total debt
/ total assets. Fixed-cost ratio = depreciation / cost of sales.

(c) p-values are the significance levels from two-tailed Wilcoxon
signed rank tests.
TABLE 3
Tests of Changes in Total Inventory Turnover and Investment (a)

                                      JIT Median   Control Median

Total Inventory Turnover
  Pre-adoption                           3.437         3.601
  Post-adoption                          4.115         3.891
  Change                                 0.407         0.073
  p-value, intra-sample change (b)       0.001         0.085

Total Inventory / Total Assets
  Pre-adoption                           0.236         0.249
  Post-adoption                          0.190         0.217
  Change                                -0.032        -0.007
  p-value, intra-sample change (b)       0.001         0.001

                                     Median Paired
                                      Difference     p-value (b)

Total Inventory Turnover
  Pre-adoption                           0.000          0.592
  Post-adoption                          0.482          0.012
  Change                                 0.401          0.001
  p-value, intra-sample change (b)

Total Inventory / Total Assets
  Pre-adoption                           0.000          0.472
  Post-adoption                         -0.022          0.002
  Change                                -0.021          0.001
  p-value, intra-sample change (b)

(a) Total inventory turnover is cost of sales divided by average
total inventory. For a given firm, pre-adoption total inventory
turnover and total inventory / total assets are median values of
the respective measures in the three years preceding the JIT
adoption year. Post-adoption total inventory turnover and total
inventory / total assets are median values of the respective
measure in the three years following the JIT adoption year.

(b) For the adopter sample, p-values for intra-sample changes are
from one-tailed signed rank tests. For the control sample, p-values
for intra-sample changes are from two-tailed signed rank tests,
p-values for pre-adoption differences are from two-tailed signed
rank tests, p-values for post-adoption differences and differences
in changes in performance are from one-tailed signed rank tests.
Tests of paired differences in changes in inventory turnover and
inventory / assets are tests of DIF[DELTA]INVTURN and
DIF[DELTA]INVASSET, as presented in Equations (1) and (2),
respectively. We conducted all analyses using 201 firm-pairs.
TABLE 4
Tests of Return on Assets Changes for JIT and Control Samples (H1)

                                               Median
                           JIT     Control     Paired        p-
                          Median    Median   Difference   value (b)

Return on Assets (a)
  Pre-adoption             0.056    0.057      0.001       0.861
  Post-adoption            0.054    0.045      0.007       0.010
  Change                  -0.005   -0.015      0.010       0.010
  p-value, intra-sample
    change (b)             0.055    0.001

(a) Pre-adoption ROA is median ROA in the three years preceding the
JIT adoption year. Post-adoption ROA is median ROA in the three
years following the JIT adoption year.

(b) For intra-sample tests, p-values are from two-tailed signed
rank tests. For paired differences, the p-value in the test of
pre-adoption ROA is from a two-tailed signed rank test, and
p-values for tests of post-adoption ROA and changes in ROA are from
one-tailed tests. The paired difference in change in ROA is
DIF[DELTA]ROA, as presented in Equation (3).
TABLE 5
Tests of Asset Turnover (H2) and Profit Margin (H3) Changes for
JIT and Control Samples

                                               Median
                           JIT     Control     Paired        p-
                          Median    Median   Difference   value (b)

Asset Turnover (a)
  Pre-adoption             1.360    1.365      0.005        0.867
  Post-adoption            1.348    1.332      0.000        0.215
  Change                   0.000   -0.018      0.035        0.033
  p-value, intra-sample
    change (b)             0.845    0.013

Profit Margin (a)
  Pre-adoption             0.040    0.039      0.001        0.825
  Post-adoption            0.038    0.030      0.004        0.010
  Change                  -0.001   -0.005      0.005        0.014
  p-value, intra-sample
    change (b)             0.114    0.001

(a) Pre-adoption asset turnover and profit margin are the median
values of the respective measures in the three years preceding the
JIT adoption year. Post-adoption asset turnover and profit margin
are the median values of the respective measures in the three years
following the JIT adoption year.

(b) For intra-sample tests, p-values are from two-tailed signed
rank tests. For paired differences, p-values in tests of
pre-adoption performance are from two-tailed signed rank tests,
and p-values in tests of post-adoption performance and changes
in performance are from one-tailed tests. Paired differences in
changes in asset turnover and profit margin are DIF[DELTA]ASSETTURN
and DIF[DELTA]MARGIN, as presented in Equations (4) and (5),
respectively.
TABLE 6
ROA Effects of JIT Adopters' Relative Profit Margin and Asset
Turnover Changes (H4) (a)

                                              Once-Reduced
                     Full Sample (b)           Sample (b)
                         n = 201                n = 176

                  Median   p-value (c)    Median   p-value (c)

MARGIN_EFFECT     0.006       0.014        0.004      0.113
TURNOVER_EFFECT   0.001       0.322        0.001      0.177
DIF_EFFECT        0.007       0.032        0.002      0.259

                        Twice-Reduced
                          Sample (b)
                           n = 162

                    Median   p-value (c)

MARGIN_EFFECT       0.009       0.015
TURNOVER_EFFECT     0.002       0.063
DIF_EFFECT          0.005       0.052

(a) We calculate the ROA effect of relative changes in profit
margin and asset turnover as follows:

MARGIN_EFFECT = DIF[DELTA]MARGIN x JIT adopter's pre-adoption asset
turnover, where DIF[DELTA]MARGIN is the paired difference in the
change in profit margin (see Equation [6]).

TURNOVER_EFFECT = DIF[DELTA]ASSETTURN x JIT adopter's pre-adoption
profit margin, where DIF[DELTA]ASSETTURN is the paired difference
in the change in asset turnover (see Equation [7]).

DIF_EFFECT = MARGIN_EFFECT - TURNOVER_EFFECT is tested to assess
the relative importance of profit margin and asset turnover
performance in JIT adopters' relative ROA changes (see Equation
[8]).

(b) The full sample is the 201 JIT adopters used in primary tests.
The once-reduced sample is identical, except that JIT adopters with
negative pre-adoption profit margins (25 firms) are removed. The
twice-reduced sample is identical to the once-reduced example,
except that JIT adopters paired with control firms with
pre-adoption losses (14 firms) are also removed.

(c) p-values are from two-tailed signed rank tests for DIF_EFFECT.
Other p-values are from one-tailed signed rank tests.
TABLE 7
Results of Two-Stage Self-Selection Analysis

Stage 1: JIT Choice Model (Equation 9) (a)

ADOPT = [[beta].sub.0] + [[beta].sub.1]PRE_SALES +
[[beta].sub.2]PRE_INVTURN + [[beta].sub.3]VAR_DEMAND
+ [[beta].sub.4]INNOVATE + u

               INTERCEPT    PRE_SALES    PRE_INVTURN

Pred Sign         (?)          (+)           (-)
Est. Coef.      -0.070       0.0001         -0.012
p-value (b)      0.740       0.050           0.734

               VAR_DEMAND    INNOVATE

Pred Sign         (-)           (+)
Est. Coef.       -0.461        4.74
p-value (b)       0.338        0.010

Model Significance: 0.001
n: 388 (194 firm-pairs)

Stage 2: Model Incorporating Selectivity Variables (Equation
10) (a)

[DELTA]ROA = [[alpha].sub.0] + [[alpha].sub.1]PRE_SALES +
[[alpha].sub.2]PRE_INVTURN + [[alpha].sub.3]VAR_DEMAND +
[[alpha].sub.4]INNOVATE + [[alpha].sub.5][M.sub.J(C)] + [epsilon]

                           INTERCEPT   PRE_SALES   PRE_INVTURN

Adopters:
  Est. Coef.                 -0.038     0.000002     -0.002
  t-statistic (b)            -0.86      2.40         -1.04
  Model [R.sup.2]: 0.027

Control:
  Est. Coef.                 -0.123    -0.00001       0.0001
  t-statistic (b)            -0.24     -0.21          0.01
  Model [R.sup.2]: 0.039

                           VAR_DEMAND   INNOVATE   [M.sub.J(C)]

Adopters:
  Est. Coef.                  0.020       0.117       -0.039
  t-statistic (b)             0.57        0.68        -0.73
  Model [R.sup.2]: 0.027

Control:
  Est. Coef.                 -0.013      -0.750        0.173
  t-statistic (b)            -0.07       -0.35         0.25
  Model [R.sup.2]: 0.039

(a) We estimated the first-stage model with probit. ADOPT equals 1
if the firm is a JIT adopter, 0 otherwise. PRE_SALES (PRE_INVTURN)
is sales (total inventory turnover) in the pre-adoption period.
VAR_DEMAND is variability of demand over the 24 quarters preceding
JIT adoption. INNOVATE is firm innovativeness, which we proxy by
R&D / Sales in the pre-adoption period. [M.sub.J] and [M.sub.C]
are selectivity variables. For adopters, [M.sub.J] is -f([beta]'Z)
/ F([[beta]'Z); for control firms, [M.sub.C] is f([beta]'Z)
+ (1 - F([beta]'Z)), where f(*) and F(*) are the density and
distribution functions, respectively, of the standard normal
distribution, and [beta]'Z is the prediction from the first-stage
probit model.

(b) p-values in the first stage are from Chi-square tests. In
testing coefficient estimates in the second stage, we used White's
(1980) heteroskedasticity-consistent estimator of the
variance-covariance matrix.
TABLE 8
Regression Results: Model Examining the Effect of Customer
Concentration (Free or Captive Status) on Relative ROA Changes

Model: DIF[DELTA]ROA = [[beta].sub.0] + [[beta].sub.1]CAPTIVE +
[[beta].sub.2]DIF_PRE_SALES + [[beta].sub.3]DIF_PRE-FCRATIO +
[epsilon]
                        Status-Matched Firm-Pairs (a)
                            (n = 114 firm-pairs)

Independent variable    Coefficient (b)    p-value (c)

Intercept                    0.011            0.059
CAPTIVE                      0.000            0.998
DIF_PRE_SALES                0.008            0.341
DIF_PRE_FCRATIO             -0.111            0.523
Regression [R.sup.2]         0.010
Intercept (without
  CAPTIVE) (d)               0.011            0.039

                         Quasi-Matched Firm-Pairs (a)
                             (n = 149 firm-pairs)

Independent variable    Coefficients (b)   p-value (c)

Intercept                    0.010            0.085
CAPTIVE                      0.010            0.346
DIF_PRE_SALES                0.005            0.531
DIF_PRE_FCRATIO              0.066            0.676
Regression [R.sup.2]         0.010
Intercept (without
  CAPTIVE) (d)               0.014            0.006

(a) CAPTIVE is coded 1 if a firm is captive, 0 if the firm is free.
The status-matched sample includes only firm-pairs matched on
customer concentration (as in BLV). The quasi-matched sample
includes the 114 firm-pairs that match on customer concentration
plus the 35 firm-pairs in which a captive adopter is matched with a
free non-adopter. In either case, CAPTIVE is coded by reference to
JIT adopters' statuses.

(b) Coefficients displayed for DIF_PRE_SALES are the actual
coefficients x 10,000. The model includes DIF_PRE_SALES and
DIF_PRE_FCRATIO to control for the significant differences in firm
size and cost structure between the JIT and control samples (see
Table 2, Panel C).

(c) Under BLV's customer concentration hypothesis, captive status
(i.e., CAPTIVE = 1) hinders a JIT adopter's post-adoption ROA
performance. Thus, negative [[beta].sub.1] supports BLV's
hypothesis. Controlling for between-sample differences in size and
cost structure, [[beta].sub.0] is the relative ROA change for free
adopters; ([[beta].sub.0] + [[beta].sub.1]) is the relative ROA
change for captive adopters. Consistent with our directional
hypothesis, intercept p-values are from one-tailed tests; other
p-values are from two-tailed tests.

(d) These intercepts and p-values are from regressions that exclude
the CAPTIVE variable.

For their helpful comments on earlier versions of the paper, we thank Ed Swanson, Bill Cready, Bob Strawser, Mike Wilkins, Tom Lopez, Robert Trezevant, Ramji Balakrishnan, Mahendra Gupta, Nick Dopuch, Bill Rankin, Jim Hesford, Bart Hamilton, Cecily Raiborn, and two anonymous referees, as well as workshop participants at Texas A&M University, Texas Christian University, Washington University in St. Louis, University of Nebraska, Western Michigan University, the 1999 American Accounting Association Southwest Regional Meeting, and the 1999 American Accounting Association Annual Meeting. We are grateful to the authors of Balakrishnan, Linsmeier, and Venkatachalam (1996) for providing us with their sample. We also gratefully acknowledge the research assistance of Jonathan Anderson, Jap Efendi, and Doug Lindauer.

Submitted May 1999

Accepted May 2001

(1) Several other studies examine the association between JIT adoption and profitability. Anyane-Ntow (1991) examines six samples of 50 JIT adopters in six industrialized nations. The author's analysis of the 1984-1986 period indicates that Japanese JIT adopters maintain lower inventory levels compared to adopters in other nations, but do not enjoy higher profitability. Huson and Nanda (1995) examine a sample of 55 firms that adopted JIT during the 1980s and find average EPS deterioration of 25 percent for adopters, and 14 percent for matched non-adopters in the four-year post-adoption period (although in a five-equation system, the reduced form estimate of the effect of JIT adoption on EPS is positive). Callen et al. (2000) examine a sample of 60 manufacturing plants that adopted JIT between 1985 and 1989, and conclude that JIT is associated with lower costs and higher profits. Biggart (2000) examines a sample of 95 finns that adopted JIT between 1975 and 1995, and reports that changes in ROA were no different for JIT adopters and non-adopting industry counterparts. Finally, Fullerton and McWatters (2001) examine JIT's profitability effects using publicly available financial data combined with a proprietary survey of 253 firms, 95 of which were pre-1995 JIT adopters. The authors document a positive relation between profitability and JIT implementation.

(2) The effect of JIT implementation on asset turnover depends on how the firm uses any capital released via inventory reductions. For example, JIT adopters that substitute noninventory capital investment (e.g., manufacturing cells) for inventory investment may reduce or eliminate any otherwise observable improvement in total asset turnover.

(3) We determined "acceptable" firm-size matches by judgment rather than by a predetermined criterion. We did not use a predetermined criterion because (1) the firm size/industry trade-off is difficult to quantify; (2) it is not clear that any predetermined firm size screening criterion (e.g., the ratio of adopter size to non-adopter size) would be equally appropriate for large and small adopters; and (3) controlling for firm size differences in statistical tests is straightforward. When we elected to match at a given industry level, we selected the control firm with net sales (in year -1) closest to the net sales of the adopting firm. The matching procedure yielded approximately 69 percent four-digit industry matches, 14 percent three-digit matches, and 17 percent two-digit matches.

(4) We do not winsorize variables that are simply size measures (e.g., total assets).

(5) Later tables reporting test results present only medians and the results of Wilcoxon signed rank tests. However, means and related t-tests produced identical inferences.

(6) We also tested raw material, work-in-process, and finished goods inventory components. In the JIT sample, turnover increased for all three components (p [less than or equal to] 0.007 for each component). Moreover, raw materials and work-in-process inventory turnover increased more for adopters than for control firms (p = 0.030 and 0.046, respectively).

(7) Operating income yielded identical inferences.

(8) As in BLV, macroeconomic factors likely account for within-sample ROA declines. In BLV, the adopter (control) sample experienced a 14.3 percent (36.2 percent) pre- to post-adoption ROA decline.

(9) We examined the extent to which firms with pre-adoption losses affect our results. We first deleted firm-pairs in which the adopters reported pre-adoption losses (n = 25). Hypothesis 1 was supported at the 0.069 level. We then deleted (in addition to these 25 firm-pairs) firm-pairs in which the non-adopters reported pre-adoption losses (n = 14). In the remaining sample of 162 firm-pairs, H1 was supported at the 0.008 level.

(10) Additional electronic searches indicated that through year +6, only six firms in our control sample ultimately reported implementation of JIT practices.

(11) After deleting firm-pairs for which adopters report pre-adoption losses (n = 25), the asset turnover and profit margin hypotheses are supported at the 0.052 and 0.125 levels, respectively. When we also delete firm-pairs with non-adopters reporting such losses (n = 14), the asset turnover and profit margin hypotheses are supported at the 0.012 and 0.015 levels, respectively.

(12) Maddala (1977, 351-366; 1983, 257-290) explains the theoretical basis for the procedure. Our study uses mean [[DELTA]ROA.sub.J] - [[DELTA]ROA.sub.C]], the mean paired difference in change in ROA from our sample, as an estimate of what we are really interested in, [[mu].sub.J] - [[mu].sub.C], which is the mean ROA if all firms adopt JIT, minus the mean ROA if no firms adopt JIT. Suppose that JIT adopters are simply superior to non-adopters, even without adopting JIT. If so, mean [[DELTA]ROA.sub.J] is an upward-biased estimate of [[mu].sub.J] (i.e., if non-adopters had adopted JIT and were in our adopter sample, mean [[DELTA]ROA.sub.J] would be smaller), whereas mean [[DELTA][ROA.sub.c] is a downward-biased estimate of [[mu].sub.C] (i.e., if adopters had not adopted JIT, but were instead in our control sample, mean [[DELTA]AROA.sub.C] would be larger). As a result, mean [[DELTA]AROA.sub.J] - [[DELTA]ROA.sub.C]] would be an upward-biased estimate of [[mu].sub.J] - [[mu].sub.C]. We use Maddala's two-stage procedure to discern whether selection bias is influencing our inferences in this manner.

(13) This data requirement reduced the number of firm-pairs in this analysis to 194.

(14) In additional sensitivity analyses related to the ROA results reported in Table 4, we regressed DIF[DELTA]ROA on paired differences in pre-adoption sales and paired differences in pre-adoption fixed-cost ratios. The coefficients on the independent variables had p-values of 0.656 and 0.086, respectively. The intercept of the equation was 0.011 (p = 0.015), indicating that the relative ROA increase for JIT adopters is still significant after controlling for the two significant between-sample differences identified in Table 2, Panel C. We also screened from the sample any firm-pair in which |adopter sales -- non-adopter sales| / non-adopter sales in the pre-adoption period exceeded 2.5. For the remaining sample of 148 firm-pairs, the average paired difference in pre-adoption sales was not significant, and the test of DIF[DELTA]ROA was significant at p = 0.020. We also examined financial statement footnotes and found that in the adoption year, 29 (17) JIT adopters (non-adopters) reported LIFO liquidation income. We re-ran the test of DIF[DELTA]ROA after reducing firms' reported earnings by the effect of such income. Our inferences remained unchanged. We conducted the following check of our inferences' validity. We regressed DIF[DELTA]ROA on DIF[DELTA]INVTURN. The coefficient on DIF[DELTA]INVTURN was positive and significant at the 0.001 level (tests using, in separate equations, paired differences in changes in raw materials, work-in-process, and finished goods turnover as independent variables yielded coefficients with p-values of 0.376, 0.024, and 0.035, respectively). Finally, we note that our H2-H4 inferences are also robust to sensitivity analyses similar to those described here for ROA. The results did suggest, however, that relative asset turnover improvement was significantly greater for adopters with relatively large fixed-cost ratios.

(15) In light of the insignificance of CAPTIVE, we re-estimated the model with CAPTIVE omitted. Near the bottom of Table 8, we report intercepts (and p-values) for the reduced model. Like the ROA results reported in Table 8, additional analyses provided no evidence that customer concentration played a role in JIT adopters' relative asset turnover and profit margin changes.

(16) For each of these four reduced samples, the p-value in the test of DIF[DELTA]MARGIN is less than 0.05. However, p-values in the four tests of DIF[DELTA]ASSETTURN range from 0.113 to 0.349. We also note that our H1-H4 inferences are unchanged when we restrict our sample to firms in two-digit industry codes represented in BLV's sample.

(17) Most of this difference is attributable to the span-of-operations screen. When we employ the screen, our sample includes 89 adopters with median pre-adoption sales of $165.33 million.

(18) To examine the effects of fixed-cost ratio differences, we conducted analyses similar to those described in this section for adopter size. We concluded from the analyses that fixed-cost ratio differences are not a factor in the two studies' different sample-wide results.

(19) Of BLV's 46 adopters, 11 (23.9 percent) report pre-adoption sales less than $21.5 million. In our sample, only 3.5 percent of adopters report pre-adoption sales less than $21.5 million. We make no claim that our sample, compared to BLV's, is more representative of the underlying population. We constructed a benchmark sample using all Compustat firms in one-digit SICs 2, 3, and 5 that reported data sufficient to calculate BLV's test variables in each year from 1984-1990 (corresponding, approximately, to BLV's data availability requirements). 1,854 Compustat firms met these conditions. Using 1988 (approximately BLV's weighted-average adoption year) as the benchmark year, these 1,854 firms report median sales of $166.2 million. The median sales value for BLV's 92 sample firms is at the 43rd percentile of the benchmark sample; the median sales value in our sample is at the 61st percentile of the benchmark sample.

(20) For all four variables of interest (DIF[DELTA]ROA, DIF[DELTA]ASSETTURN, DIF[DELTA]MARGIN, and DIF_EFFECT), multiple regression models estimated with our sample of 201 firm-pairs yield no evidence of statistically significant effects related to adopter size or adoption timing, or to an interaction thereof.

(21) We cannot definitively attribute our H4 findings to either reduced inventory holding costs or reduced production costs. It seems unlikely, however, that reduced holding costs alone would drive our sample-wide finding that profit margin dominates asset turnover in explaining JIT adopters' relatively improved ROAs.

REFERENCES

Alles, M., S. M. Datar, and R. A. Lambert. 1995. Moral hazard and management control in just-in-time settings. Journal of Accounting Research 33: 177-204.

Anyane-Ntow, K. 1991. Just-in-time manufacturing systems and inventory reported in financial statements: A cross-national comparison of manufacturing firms. The International Journal of Accounting 26: 277-285.

Atkinson, A. A., R. D. Banker, R. S. Kaplan, and S. M. Young. 2001. Management Accounting. 3rd edition. Upper Saddle River, NJ: Prentice Hall.

Balakrishnan, R., T. J. Linsmeier, and M. Venkatachalam. 1996. Financial benefits from JIT adoption: Effects of customer concentration and cost structure. The Accounting Review 71 (April): 183-205.

Biggart, T. B. 2000. The effects of just-in-time inventory system adoption on firm performance. Working paper, University of North Carolina at Greensboro.

Callen, J. L., C. Fader, and I. Krinsky. 2000. Just-in-time: A cross-sectional plant analysis. International Journal of Production Economics 63: 277-301.

Fullerton, R. R., and C. S. McWatters. 2001. An empirical investigation of the relationship between JIT and financial performance. Working paper, Utah State University.

Hogan, C. E. 1997. Costs and benefits of audit quality in the IPO market: A self-selection analysis. The Accounting Review (January): 67-86.

Huson, M., and D. Nanda. 1995. The impact of just-in-time manufacturing on firm performance in the U.S. Journal of Operations Management 12: 297-310.

Lubben, R. T. 1988. Just-In-Time Manufacturing: An Aggressive Manufacturing Strategy. New York, NY: McGraw-Hill.

Maddala, G. S. 1977. Self-selectivity problems in econometric models. In Applications of Statistics, edited by P. R. Krishnaiah. New York, NY: Elsevier Science.

--. 1983. Limited Dependent and Qualitative Variables in Econometrics. Econometric Society Monographs in Quantitative Economics No. 3. New York, NY: Cambridge University Press.

--. 1991. A perspective on the use of limited-dependent and qualitative variables models in accounting research. The Accounting Review 66 (October): 788-807.

McIlhattan, R. D. 1987. How cost management can support the JIT philosophy. Management Accounting 69 (September): 20-27.

White, H. 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48: 817-838.

Young, S. M., and E H. Selto. 1991. New manufacturing practices and cost management: A review of the literature and directions for new research. Journal of Accounting Literature 10: 265-297.

Michael R. Kinney
Texas A&M University
William F. Wempe
Texas Christian University