Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Comparison of Industry Output Measures in Manufacturing JOEL POPKIN Joel Popkin and Company INTRODUCTION When in 1962 the Bureau of Economic Analysis (BEA) introduced annual measures of gross output originating by industry, the possibility arose that differences in behavior would emerge between them and the Federal Reserve Board's (FRB) industrial production indexes, which were first published in 1926. The two series have in fact moved differently, which has concerned producers as well as users of the data. Year-to-year and trend differences can be seen when the two series are compared. Year-to-year differences have been of concern recently because of speculation that the process of deriving constant-dollar series from the nominal value of shipments and the book value of inventories may be par- ticularly subject to error in a period of high inflation rates and of shifts in inventory valuation methods such high rates induce. Such concern reached its peak during the recent bout of double-digit inflation. Trend differences in the movement of the two series have also received attention as a result of concern over the apparent decline in the rate of productivity advance in the past 10 years. It has been about a dozen years since the last study of the two series was accomplished. It was prepared and presented by Gottsegen and Ziemer (1968) at an Income and Wealth Conference in 1964, the proceedings of which were published in 1968. The study dealt with differences arising between the two series due to both concepts and estimation methods, and covered the years 1947-1964. The aggregate output, as measured by FRO' 363
364 PAPERS rose somewhat faster than that of BEA during those years, when the underlying data are standardized with respect to concept. Since then, and particularly after 1967, the gap between FRB and BEA has continued and widened. It is apparent in a similar analysis published by the FRB in connection with its 1971 revision of the industrial produc- tion index. This analysis covered the years from 1947 through 1971. Discussions with officials who prepare these indexes indicate the divergences widened further after 1971, and there was a particularly marked difference in 1974, a year of double-digit inflation. Much time has passed since the Gottsegen-Ziemer study. The purpose of this research paper will be to carry forward the investigation of the sources of differences in the movement of industrial production and real output originating in the industrial sector. The circumstances of declining trend productivity and high inflation rates, double-digit at some times, create the need for such an update. The main thrust of this paper is to examine both the FRB and BEA out- put data in some detail and attempt to understand the reasons for the divergence such as it has been. More attention will be given to statistical procedures as a source of divergence than to concept. The focus of the project is the comparison of BEA'S annual measures of the deflated value of output for the 51 manufacturing cells of the 80-order U.S. input-output table with similar measures constructed from the FRB production index system from 1963 through 1974. The time period is selected to start with a benchmark year and run through one full and one partial benchmark year, 1967 and 1972. Thus, it begins about when the Gottsegen-Ziemer study ended, to avoid duplication, and goes beyond the 1972 partial benchmark through 1974 in order to provide an assessment of the effect of the high inflation rates of 1973-1974 on the measurement of output. THE DATA SET AND TIME PERIODS To facilitate a comprehensive review and analysis of the behavior of the two series, the data set selected should be as disaggregated as possible. The FRB prepares 215 basic series for manufacturing, at a degree of detail averaging somewhat greater than 3-digit sac. But BEA prepares output measures only at the level of disaggregation of the 51 manufacturing cells of the 80-order U.S. input-output (I-O) table, on average between 2- and 3-digit detail. The degree of detail available from BEA iS determined by the level at which it calculates the change in goods-in-process and finished goods inventories necessary to transform its deflated shipments data into
Comparison of Industry Output Measures in Manufacturing 365 gross output measures. The BEA uses these output measures to derive deflators for gross product originating calculated from the income side of the accounts; but for maximum consistency in comparing FRB and BEA data and the growing use in sectoral research of gross output rather than value added, the analysis here is based on output-side measures. The BEA provided annual data on gross output for the S1 cells for the years 1963-1974. The terminal year was dictated by the availability of data on a 1967 sac definitional basis. FRB data are currently defined on that basis. A concordance was prepared by which the 215 FRB series could be ag- gregated to correspond with the classifications used in the 1967 I-O table. FRB concorded with BEA at that level except in one instance in which FRB data cut across two I-O cells. These two cells (62 and 63) had to be col- lapsed. So the data set was reduced to SO sectors. The FRB prepared an- nual data for these 50 sectors for the same time period. The two data sets are not strictly comparable in concepts for a number of reasons. Probably most important is that FRB data were aggregated to the 50 sectors using value-added weights because gross output value weights were not available. But given the sectors employed in the study, there is not a wide variation in the ratio of value added to gross output. Another noteworthy difference is that the FRB indexes are based on 1963 weights from 1963 to 1966, 1967 weights thereafter. The BEA output measures are based largely on Bureau of Labor Statistics (B[S) price in- dexes, which are weighted by 1963 data. These factors, related to weights, would loom troublesome if the analytical findings were of small dif- ferences between the two series. But they are not important enough to give rise to widespread, large differences between years. To conduct an analysis of differences between the two series, it is necessary to divide the annual data covering 1963- 1974 into several subperiods, as well as to treat the period in its entirety. The need for subperiod analysis reflects FRB'S use of different data sources for different time spans. To measure changes between benchmark years, which coin- cide with the roughly quinquennial censuses of manufactures, FRB relies in large part on 6,000 or so production indexes calculated by the Census Bureau. These indexes are based on physical quantity measures except where they are not available or deemed inadequate. Where physical quan- tities are not the basis for the indexes, they are derived by adjusting shipments for inventory (goods-in-process and finished goods) change and deflating the resulting value of production. Deflation is based on specification-based price indexes of the BES or on unit-value indexes developed by the Census Bureau. The criteria for selection of the ap
366 PAPERS propriate price series have changed over the years, sometimes swinging in the direction of the BUS indexes, sometimes, the unit values. As noted, the FRB does not use the Census Bureau indexes ubiquitously. Its description of the 1976 revision of the production indexes contains the following: . . . the food, petroleum refining and primary metals industries were some of the major areas where benchmark and annual production levels for a number of series were based on averages of monthly physical quantity data. Special compilations of data made available by trade associations and other industry sources were the basis for new production levels in some industries, including fabricated structural steel and aircraft. In addition, alternative production indexes for certain portions of the textile, printing and machinery industries were compiled by the staff of the Federal Reserve Board of Governors. The changes between 1963 and 1967 in the production indexes for the 50 sectors on which the analysis is based were calculated by FRB in the manner indicated above. The annual changes between those years were in- terpolated primarily by deflated value of production data from the Census Bureau's annual surveys of manufactures. These data are more ag- gregative than those on which the intercensal-year production indexes are based. And the FRB uses measures alternative to those of the Census Bureau in the areas indicated above, where it has eschewed use of the Cen- sus production indexes. Changes during 1963-1967 constitute one subset of data that requires separate analysis. Between 1967 and 1972, another year for which a quinquennial census of manufactures was taken, the FRB has completed a "partial" bench- mark revision. The revision is termed partial because it does not yet incorporate the detailed Census production indexes. It is limited in manufacturing to expansion of the number of series, and use of more comprehensive information for measuring output in petroleum refining, chemicals, construction supplies, and motor vehicles. The revision also in- corporates throughout the index components more comprehensive annual data through 1973 and more accurate monthly series for interpolation through 1973, extrapolation thereafter. Because the data for 1967-1972 reflect partial revision and will be revised later to reflect a complete bench- marking, this subperiod will be analyzed separately. A final subperiod is that of 1973-1974. As noted, some revisions have been made in data for 1973 and in the monthly measures used to extrap- olate 1974. These two years form a subset not because of FRB methods of calculation, but rather because of interest in comparing direct output measures with those based on deflation during a time period when the deflators were accelerating rapidly.
Comparison of Industry Output Measures in Manufacturing 367 ANALYSIS The analysis that follows is divided into four parts in line with the forgo- ing discussion of how FRB constructs its production indexes. The first part is concerned with the analysis of changes between 1963 and 1967, full benchmark years. The analysis addresses both average annual changes be- tween the two benchmark years and changes between individual years, 1963-1967, for the 50 industries. The second part of the analysis deals with changes between 1967 and 1972. The average annual changes be- tween these two years and between the intervening years are analyzed. The third part of the analysis deals with the 1973-1974 period. Here the analysis is based only on year-to-year changes because benchmarking does not enter at all. The final section deals with the results of analysis covering all years, 1963-1974. The data underlying these analyses are found in the appendix. ANALYSIS FOR 1963-1967 In Table 1 are found average annual rates of change for the 50 industry sectors from 1963 to 1967 for both the FRB and BEA series. The difference between the two series is also shown. For the FRB series, differences in the average annual rates of growth range from 7/10 of 1 percent for tobacco manufactures to 27.0 percent for electronic components and accessories. The simple average annual rate of increase of the FRB series is 7.6 percent. For the BEA series the slowest and fastest rates of growth are registered for the same two industries, although the BEA growth rate for tobacco is slightly negative and that for electronics is considerably less than the FRB index shows. For the entire period the BEA measures average 7.1 percent across all 50 sectors. Thus, the difference between the two series is 0.5 per- cent. Table 2 contains a number of regressions that were estimated using the two sets of data. The general form of the regression was to specify percent- age change in the FRB index as a function of the percentage change in the BEA gross output measure. Estimates were obtained of the intercept and slope of that relationship. Obviously, if the two measures were identical, the intercept would be zero, the slope one, and the correlation 1.00. The first regression in Table 2 is that of the average annual rates of change be- tween 1963 and 1967 for the FRB series and for the BEA series across all 50 industries. As can be seen, the slope term is 1.086, somewhat greater though not significantly larger than unity. The intercept is not signif~- cantly different from zero. In general, the observations are fairly well cor- related; the adjusted coefficient of determination is 0.88. The regression
so .= ~ of C,2 ~ 'e - ~ ~ As by o ~ cat au cd x - ~ If - - "s: ~ ~ o a, . - by - ~ 'e 3 ~ C) ._ Cal ~ o C) - ~s o c: ~ Cal ~ - CQ Ct ~ =: En Ct Ct no ~ O Ct X . c: - , . ° O , i 1 so ~ o Ct oo e o c: - _ ,_ ~4 Cal ~ o V] - o Cal a' ~ <: m am m m o~ - ~ ¢ m ~m m a~ oo m C ~ _ ~ m m r ~m _ m ° ~ ~ - U Z _ .- ~S _ U' - 1 0 ~ ~ . . . C~ ~ ~ . . . ~ _ _ ~ ~ ~ oo _ ~ _ C~ oo ~ oo ~ oo 4 O O O - O O ~1 1 1 ~ C~ C~ ~_ ~ ~ . . . .. . . ~ 00 ~ ~ U~ C~ 1 O - C~ ~)~1' 00 ~ . . . .. . . ~ ~ ~ ~ U~ ~ - . . . .. . . 00 ~ ~ ~ oo ~0 _ ~ ~r _ _ ~ _ _ _ ~ ~ C~ C~ ~ oO O ~ ~ ~ ~ r~ ~c~ ro ~c~ ~ _ . . . . . . O O - O O OO - O O ~ - O O ~O - 1 1 11 1 00 _ ~ ~ 00 ~ ~ ~ ~ . . . . . .. . . . ~ O ~ 0~ ~ ~ _ U: ~ 1 ~ ~ ~ ~ ~ r~_ ~ _ ~ . . . . . .. . . . ~ 0 ro oo ~ ~ ~ ~ u~ 1 ~ 00 ~ ~ ~ _O ~ O _ O O ~ O - OO - O O 1 1 11 1 ~ ~ ~ ~ Cr~ ~ . . . . . . ~ O ~ 00 _ ~ ~ O . . O U: a~ ~ ~ o ~ o ~ ~ ~ ~oo - ~ c ~ ~ - ~ u~ o o oo o ~ ~u) u) ~r~ oo ~ ~ c~ c~ 3 K O K ~v' :t o (L' ~K 3 V, t~ ro ~ ·- 3 = _ ~ ~ ~ ~ ' ~ _ 368
- ~_ ~ ~ ~ cr' ~ ~0 ~ ~ cr us _ ~0 ~ ~ or ~or or - or or - or - ~ or ~ or or or or - - or ~ or 1 1 1 1 1 1 1 ) ~ - - o - o ~ ~ - ~ ~ o ~ ~ ~ 1 1 oo '=~ - o oo o c~ o - - oo _ ~ cr ~ ~ ~ ~ ~ ~D ~oO ~ ~CN - ) ~ ~ uD ~ ~ ~ o ~ ~ - 1 1 - - - ~O oo ~ ~ - Lr) o - - ~ o ~ - ~u: o o ~ ~oo o o ~o o c~ o c~ - o ~ ~ o ~ o o o - ~o - ~ o 1 1 1 1 1 1 Lr) c~ ~ c~ 0 ~ ~ ~ ~ r~ 0 ~_ ~0 0 ~ _ 1 1 1 1 1 1 1 1 1 1 1 o~ ~ ~o o 11 ~- ~ oo o~ ~c~ o 11 o ~ oO ~O _ C~ r~ ~oO a~ 0 - ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ u ~ ~ s~ c~ c<S ~O CL~;0 ~ ~ D~ 3 4=,,, ~0 ~, .S S~, =-o a , & · ' u} ' ~ ~ _ · 369 ~4 ~: s: . ~C~ - ·- ~C~ ~.= £ ·- ~3 ~C) 0 ~c, £ ~50 0~0 £ ~£ .= ;:-- =.= .' ~ E o E.' =-' E 0, °.' ' ·= ·= 3 ·= <e ~ ~ ·~t ~ c' ce .c £ ~ ~ ~o ~ ~ ~ V ~ ~ ~ '
c~ - 1 r~ a~ - a~ - 1 oo - ~D - 1 ~D - ¢ 1 <: : uO~ o - ;G ~t^~ 1 ~5 :q x ~: 1 ~: LLI ~: 5 _ ~ O ~ ~ _ C: Z; ~ ._ _ ~_ ~ _ _ O _ V) <= V' ~ _ 1 ~ ~C~ ~ C~ U~o o .. .. . . .. . OO ~O O O ~_ O 1 11 O ~ ~ U~oo O . . . .. . ~ ~ 0 a~ ~ 1 u:r~ OC~ O ~ .. . . r~ ~ 0 -1 OoO O .. . oo~ ~ o ~_U~ _o ~ ~ ooU~ _ . ... .. . . .. . _ ~o~ ~O O O __ O 1 -. . .. _ ~ ~ ,~ ~ _ _C~ O . .. . . .. . ~ ~U) ~ oo oo,L~ U~ ~ oo ~_ ~ ~ ~ ~ ~ . ... .. . . .. . c~ ~0 a~uO ~ ~ ~r~ ~ 1 r~ ~O- c~ ~ O ~ ~ . ... .. . . .. . O O-~ ~O _ _ _O O ·r) ~00 ~ . ... . ~0~ O r~ oo ~O~ O . ... . ~ ~0 ~oo U: U: oo _ oo ~_ ~ . . . . . . ~ C~ O _ _ ~ _ _ _ ~ ~ oo _ oO - . . . . . . ~ ~ a~ 0 0 ~ _ _ r~ O _ ~ ~r U~ V7 . _ ~V' ._ ~ 0 -1- . ~ · .= 3 't ~ ~ ~ =-= ~ W -C~ ~o - oR ~ ~ _ U L~ ~ ~ ~ ~ ~ ~ ~ ·~- ~ ~ .~ ~ ~ o ,' 370
to c~ s~ c~ ~; o . - o · c'7 ~ ~ ~ c~ u) u) ~ ~ o 5 . O Z O Z~1 ~ . ~ 6 C~ C~ o CQ ~.~ -, .C-t ~ _ Q.~ ~ O o ._ C) o : - ._ C~ Ct t4 C~ ~ _ 1 C) ~ _ ~_ V O C~ ~ ~V O ~ ._ U. C~ - C~ ' ,~ ._ r ~_ 00 a~ - _oo ~_ ~ _ ~oo _ r~ OC~oo U-)C~- 1 ~a~ ... OOO oo C ~oo ~ CNr ~ __ oo ~ ~ -~r . . . .. O O O OO o __ ~_O r~ ~ - ~- ..... . . ... O ~oo~ ~ oO ~oo _ 1 ~Oo ~oo ~ - O r~- ~oou~ _ 1 oo0 oo~ OO s~ au O ~ ' · c,~ ~ ~L C~ :- O ~L~ Ct ~ O C) S:L - V) Ct - ~! ~ O c) C~ U) ~ _ ~0~ .. oo~ ~U: u~Ocr ... _ ~O ~ ~ C~ ~C~ r~oooO r~ ~ ~oo ... . . ... OOO O O O,OO _0 ~ r~ r~- x ~ - .. . . ... _- r~ ~ r ~- ~r ~- ~ ~r r ~ O-OO- O O -OO ..... . . ... OOOOO O O OOO oO oO ~) ~oOx) oo-- oo X ~oooO O-- O O OOO ooC~ oo r ~ r~ ..... . . O-OOO o O O ~OO ~ ~ _r~O ~- ~ 00 =. . O O - O O - - ~ c~, 1 1 1 1 O ~o .. . OO O oo~ .. __ 1 ~: ~ ~' ~ A Vl o ~- ~ o ~ ~- ~ o~' ~ = g ~ C,) ~- ~ O~ ~- Ct _ ~ _ ~ ~ - c~c ~V ~V~ c~l~V: 0 0 5 ccl L c ol L 0 0 V 4 3 C "D ~ ~ ~ ·' 't A -- V '= ~ 't o~ -= - -, - ~ - °= ~Q) C~ ~ ~ V ~ C) ~ C) g C=, ,,, C,~, g ~ ~ 6 ~ 6 L 6 L 6 ~ 6 ~ 6 6 ~ ~ ~ ~ - 'J <) . . . . . . . . . . =N _ ~ ~ ~ 30 G ~O _ _ 371
~ 5 ~ ce V~ V~ ~ ~ o , ._ O O ~ Z O ~ s:: ~m .~: m Ct << c4 ~ m Ct ~ m .. O. .= ~ ~ S~ c: 0 ~ ~ 0 _) ~ ~ ~ S~ o~ ~ ~ 0 ~ 0 ~ S~ ·C~ V) Ct 50 C) ~ 0 Ct S..d 0 V, S~ Ct 5_ C: ° ~V Ct V) ~i ~ C} CD _ C~ ~; - ·_ ~_ 372 ~U) ~ - o ~ . ~o _ o _ Lr ~ _ ~_oo o ...... . o_oo_o _ ~_ ~_ . . . . . . . U) o oo . . ~ a, .. . ~r~ ~ _ oo ~oooo ...... . oooooo o _a~oo a ~ ... ___ oU~ o ~oo ... _oo ~D O o ~_ - oo ~ ~ o_o o ... . ooo o o o _ ~ ~ . . o o - oo ~ o Oro ~u~ ...... . o ~oooo o o . - ct - oo - - ~o c~ c`S · s" c) <¢ ~ ~u ~ r ~ - o o ~- ~ c) ~c) ~: e 't 'e " <~ I m c~ m ~1~ ~J° V c' m c' m P: <: ct ~: c' ,,= ce c~ c} c o~ 04 s: ct ~ c~S ~ ~s ~ ~ v, c~ ~ c.) s: ~ co c) 04 ~ ~ ct ~ ct O s~ ~ 5- ~ c~ O c' ~ 0 ~ U ~ _ .. . . ~ L~ _ _ _ _ ~ ~ ~ ~: ct c~ <: o A m ~c~ t4 o~ ct ct c ~c c ~_ s ~ ~ ¢ . . _ _
\O uO~ ~ ~ u ) oo ~r~ ~ ~ o~ U) ~_ u~ u~ ~r~ ~ r~ - ~ ~ - - c ~ - ~ oo - ~o~ ... 1 ~uO ... ~ro~ u ~O r~ oo ~ oO ~ O O oo o ~ ~- - ~ Lo c~ (~) - ~ ~ - oo ~ ~ . . . - ~ u) . . . ~ - - 1 _ . . . .. ~ . ~ ~ ~ ~- - - ~ u~ ~ ~ ~ ~ . . . .. . . r~ O O ~ ~ ~ 1 1 1 ~ ~ oo~ ~ ~ r~ u~ oO . . .. . . .. . . ~ oo ~ ~ oo ~u: oo ~ '1 1 -oo-~ ~ ~oo ~ - o ~ c~ ~ r~oo~ ~ 0u~ u~ ~ ~ ~ oooo o oo o o o o o o c-] \= ~-oo~ ~ ~ ~ ~ ~ ~ ~ ~c~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ooo - ~ ~(~4 (~) tc~ oooo o -o - - o o o o .... . .. . . . . . . oo.oo o oo o o o o o o r~ aocr~ ~ _ .... . . oooo o o 1 o ) ... ooo ~a~ u) ... --- 0 ~ ~ cr ~ ) ~ oo . . .. . . . o o ~- - - o ~ oo oo~ c~ ~ ~ oo 0r~ . . .. . . . ~ ~ ~ ~ oo U) 1 ~ U~ oo . . . O 0 0 ~ ~ oo . . . O 0 0 ~o ~ 0 a~ oo 0 . . . 1 0 ~0 0 (e 0 0 Vl ~ ~A Vl .~_ ~ ~A Vl ~ a: ~;-~ ~ x ~ ~ ~ ~ o ~ ~ ~- ~ ~ ~3 ~ · ~_ ~ ~ ~ ~ C13 ~ C~ C) ~ ~ ~ ~C) C) ~ ~ 04 t4 ~o oc o~ ~ ~o.o t.4 oc o.o oc C: ~ ~ ~ ~s: s: ~ ~ ~ ~ s: ct ~ct ct ct Cect ct ct ct ct ct Ct ~S ~ ~ ,=~ ~S ~S C) C~ ~ t> C) C) C) C~ ~ ~ C) C) C) C) ct ct v~ ct C) ~; ~ce 5~ ~t ~ ~t ~ ~t ~t ~t ~ · ~ 3 ~ s ~ ~ 8 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ", ~ ~ ~ < ~ ¢ _ ~ c, < ~ ~ ~ < ~ ~ ~ ~ ~ ~ ~ ~ a~ 0 ~ _ ~ ro ~ ~ ~ ~ 0N 00 a~ 0 _ _ ~ _ c~ c~ ~ c~ ~ ~ ~ ~ ~ ~ ~ 373
374 PAPERS results suggest that there is a small but increasing divergence between the two series, FRB minus BEA positive and increasing over the range of obser- vation. The sixth regression in Table 2 covers the same time period, but each of the four annual observations of the percentage change is used; they are not averaged over the 1963-1967 period as in the first regression. Perhaps surprisingly, the results of this regression differ in important respects from the first regression. The R2 falls, but that is to be expected given the usual increase in noise as series are disaggregated, either with respect to sectoral detail or time period. The interesting aspect of the difference is that in regression 6 the intercept is positive and significantly different from zero and the slope is significantly less than unity. This suggests that there is an aspect in which the annual interpolation of data be- tween benchmark years has a different structure than the overall change measured between benchmark years. To explore this result further, two additional regressions (7 and 8) were run. In regression 7, only those observations in which the FRB index showed ~ positive percentage change were used. In regression 8, all the remaining observations, those in which FRB was unchanged or declined, were included. The difference between the results of these two equations illustrates a property that seems to exist throughout the entire time period of analysis, as will become clear as subsequent results are presented. Regressions 7 and 8 are significantly different from each other as determined by the Chow test. For regression 7 the intercept is positive and significant, and the slope is significantly less than unity. For regression 8 the intercept is negative and significant, while the slope, which is significant, is nearly half that found in regression 7. This result suggests that in the interpola- tion of annual changes between benchmark years, the positive differences between benchmark years get reflected annually in those series that are in- creasing on an annual basis. For such observations, the average difference between FRB and BEA exceeds that found on average to exist for the 50 series between benchmark years. A partial offset is provided, however, from those series or observations that show a decline on an annual basis; BEA tends to exceed FRB in those cases. This result seemed to suggest a possible reason for the discrepancy found between FRB and BEA between benchmark years; it might be at- tributable to those industry series in which FRB rose between benchmark years. Unfortunately, for this analysis, just about all series for the SO sec- tors showed a positive gain on average between 1963 and 1967, so there were too few negative observations on FRB series to split the 50-observation sample into two components. Instead, a dichotomy was made between those of the 50 series that showed an average increase in excess of the
Comparison of Industry Output Measures in Manufacturing 375 mean for the 50 series and those that showed increases less than the mean. Results are shown in equations 2 and 3 in Table 2. The structure of these two regressions does not differ significantly from that of regression 1 so there is nothing to suggest that between benchmark years, positive dif- ferences between FRB and BEA arise from the measurement of those series in which FRB iS rising more than the average for all series. Before leaving the analysis of the years 1963-1967, one more test seems appropriate. As mentioned above, the FOB does not use Census produc- tion indexes in a number of industries, accounting for about 25 percent of the weight of the total production index. Of the ~ industries, the measurement of change between benchmark years in 17 is based all or in part on measures developed outside of the scope of the Census Bureau's production index. Accordingly, the sample of benchmark changes be- tween 1963 and 1967 was divided into two parts, one consisting of these 17 industries, the other of the remaining 33 industries. The results are presented as regressions 7 and 8, which do not differ significantly from regression 1. Similar results are found in regressions 9 and 10 based on annual observations from 1963 to 1967. Again, there is no significant dif- ference between the two equations and their parent. The results of analyzing the behavior of FRB and BEA measures of out- put change during 1963-1967 can be summarized at this point. First, with respect to average rates of growth between those two benchmark years, the hypothesis cannot be rejected that the relationship between the two series is characterized by a regression with an intercept of zero and slope of one, namely, that there is no significant difference between the two series with respect to the rates of growth they show for the SO industries between the two benchmark years. Furthermore, when the industries are divided, with respect, first, to whether or not they are based wholly on Census produc- tion indexes and, second, whether their FRB output indicators are rising more rapidly or less rapidly than the average for all industries, the null hypothesis that the regressions based on these data subsets are not dif- ferent from their aggregate cannot be rejected. When annual changes between benchmark years are subject to similar regression analysis, tests that shed light on the way in which changes be- tween benchmark years are interpolated, the results suggest that the size and algebraic sign of the difference between FRB and BEA are influenced by whether or not the FRB measure is rising. ~ ^ ANALYSIS FOR 1968-1972 Average annual rates of change for the two series between 1968 and 1972 are found in Table 1. The difference between the simple average rates of
376 PAPERS change of the two series is 0.7 percent, somewhat greater than for 1963-1967. And the regression relationship of the two series (regression 11 in Table 2) is different. The intercept of that equation is 0.79, which is significantly different from zero. Since the slope is not significantly dif- ferent from unity, the implication is that there is a fairly constant dif- ference of about three fourths of a percentage point per year between the series in the 1968-1972 period. The results of dividing the observations in regression 11, first, into two groups based on whether the observation shows growth greater or less than the mean (regressions 12 and 13) and, second, with respect to whether or not the industry observations are based entirely on FRB production indexes, are that neither breakdown shows significant difference from the aggregate summary given by regression 11. Regression 16 presents the results of a regression of the percentage changes in FRB on those in BEA annually for the period 1968-1972. As was the case in regression 6 for the 1964-1967 period, the interpolation an- nually of changes between benchmark years does seem to reflect a dif- ference depending on whether or not the FRB index is rising or falling. These results can be seen in regressions 17 and 18, which are structurally significantly different from regression 16. In regression 17, the regression of all annual changes during 1968-1972 for those observations in which the FRB index is rising, the intercept is quite large and significantly dif- ferent from zero, while the slope is about 2/3. In regression 18 the intercept is negative and insignificant, and the slope is about the same as that of equation 17. When the FRB index is rising, it exhibits a positive but nar- rowing difference from the BEA measure. When the FRB index is falling, the BEA measure tends to fall less than that of FRB, except when the FRB index falls by a fairly large amount. Regressions 19 and 20 indicate that whether or not FRB series are based totally on Census data makes no dif- ference in the interpolation of annual changes between 1967 and 1972. ANALYSIS OF 1973-1974 The years 1973 and 1974 were unusual years in that there were large rates of inflation and the view is widely held that deflation is fraught with poten- tial for error in such years. The data for these two years is found in the ap- pendix. Regression 21 in Table 2 reflects the relationship between annual changes in both of those two years for the 50 series. The intercept of that equation is significant and substantially larger than the positive intercept in the annual change equations for the two preceding subperiods (regres- sions 6 and 16~. The slope is somewhat less than I/: and is significant; it is about half the size of the slope of regressions 6 and 16, and the adjusted
Comparison of Industry Output Measures in Manufacturing 377 coefficient of variation of this equation is the lowest encountered so far in the analysis. When the observations in regression 21 are subdivided into those in which FRB is rising and those in which it is unchanged or falling, the same kind of significant structural difference occurs. The intercept in regres- sion 22 is the largest estimated so far, and the slope is only try; both parameter estimates are significant. For regression 23, which contains those observations for which FRB is unchanged or falling, there is virtually no correlation between the two series. Since all but three of the declining observations occur in 1974, separate regressions for 1973 and 1974 differ significantly from each other. Since there is no reason to expect that the problems attendant to deflating the value of output are correlated with the direction of change in the FRB indexes, the effects of double-digit inflation do not appear to explain the 1973-1974 results. The results for 1974 and to some extent for 1973 are based on com- pletely unbenchmarked FRB data. During that time period, FRB makes considerable use of a variety of monthly series for extrapolating output. As a result, the series at the 50-industry level are of four general types: physical quantity, kilowatt hours, labor hours, and a mixture of one or more of the three. Equations were run for each of these four subcategories of the observations for 1973-1974. The results are contained in regressions 24-27 and do not appear to differ in any important way. ANALYSIS FOR 1963-1974 Regression 28 (Table 2) summarizes the relationship between the annual changes in the BEA and FRB measures for the entire sample period. The in- tercept of the equation is significantly positive, and the slope is significant and equal to about His, When the 550 observations on which this regres- sion was based are divided into the two now familiar groups in which FRB change is either positive or not, the same general results are obtained. For positive changes the intercept is positive and significant and the slope is less than one. For negative changes, the intercept is negative and signifi- cant and the slope is significant, but smaller. While the regressions for this time period, and the others as well, pro- vide convenient summary statistics of the underlying relationships be- tween two series, the fullest picture of the relationships can be obtained only by looking at the data in the appendix. The large differences are clearly manifest in the years 1973 and 1974. In 1973 the differences be- tween FRB and BEA measures range from 22.5 percentage points for wooden containers (I-O 21) to -9.7 percentage points for paints and
378 PAPERS allied products (I-O 30~. For 1974 the range is even wider, from 31.3 percentage points for chemicals and selected chemical products (I-O 27) to-12.4 percentage points for lumber and wood products (I-O 20~; and it is of importance that the largest outliers in 1974, a year of output bot- tlenecks in primary manufacturing industries, are found in two such in- dustries. But one need not look at only primary industries and the years 1973 and 1974 to observe large year-to-year differences. In fact, some of the most persistent differences throughout the entire sample period are found in producer's goods industries. In measuring the output of electronics (I-O 57), office, computing, and accounting machinery (I-O 51) and construc- tion machinery (I-O 45), sizable differences are found in the majority of observations. This would seem to suggest that in some industries, FRB and BEA may be measuring totally different real output universes. POSSIBLE IMPLICATIONS OF FINDINGS This paper has three main findings about the behavior of FRB indexes vis- a-vis those of BEA. First, over long periods of time, FRB output indexes for manufacturing on average rise somewhat faster than those of BEA. Be- tween the benchmark years of 1963 and 1967, the average annual differen- tial, unweighted, was 0.5 percentage points. Between 1967 and the partial benchmark year of 1972, the differential was slightly larger, 0.7 percent- age points. Perhaps this difference will be reduced when 1972 FRB data are benchmarked to the Census production indexes, but the difference widened as a result of the 1967 benchmarking. Upon regression analysis, the difference during 1963-1967 is reflected in a slope term of 1.06, somewhat greater than unity, but not significantly so. The gap is greater the larger the rise. For 1967-1972 the gap is con- stant, at 0.79 percentage points, since the slope is exactly unity. Trend differences between these output measures may be attributed to a number of factors, but the results here do not support any one of them conclusively. One such factor is the handling of quality change. It is sometimes alleged that the adjustment of direct quantity measures for quality change is inadequate. A similar allegation is made frequently about price indexes used to deflate value measures in arriving at quantity measures indirectly. The direction of quality bias in direct quantity measures and price indexes has not been shown conclusively. The preponderance of conventional wisdom is that the treatment of quality change in direct quantity measures results in their downward bias and in an upward bias in price indexes. On this assumption, quantity measures
Comparison of Industry Output Measures in Manufacturing 379 would have to be biased downward by less than the price indexes' upward bias to reconcile the FRB-BEA gap between benchmark years. A second possible source of the small but persistent trend difference between FRB and BEA output measures hinges on an argument related to sampling. Since both FRB and BEA rely on sample-based statistics- FRB, quantities and BEA, prices and the variance of quality change across commodities is greater than that of the prices, then prices are more accurately sampled than quantities. However, FRB benchmarking is based on 6,000 Census production indexes plus other measures for the 25 per- cent or so of the components for which FRB uses different sources. Price indexes for manufacturing during the 1963-1974 period were probably based on average on about 2,500 price observations. So the question, which will not be answered here, is how much larger does the quantity sample have to be vis-a-vis the price sample to equalize their accuracy? A third factor introduced sometimes in explaining the FRB-BEA spread is the difficulty in measuring constant-dollar change in goods-in-process and finished goods inventories, a calculation BEA (and to some extent FRB) must make to convert deflated shipments to production. These calcula- tions, as presently made by BEA for the 50 industries, are not bench- marked to the aggregate inventory change data used in the NIPA. Even if that were done, however, as is intended by BEA, the same kind of prob- lems would remain in measuring inventory change in the NIPA. Finally, there is a whole host of inconsistencies such as weighting, sac definition, etc., that could and probably does contribute to the discrep- ancy. However, the effect of these inconsistencies is likely to be small, if not offsetting, in the aggregate. The difference in behavior of the two output series, though small an- nually, does mount and could, if resolvable, shed some light on the re- cently observed slowdown in productivity. But the areas for research. quality adjustment, sampling, and inventory measurement, will absorb considerable resources and may well not be conclusive with respect to the trend differences. Perhaps more promising, both per se and vis-ri-vis understanding the trend differences, is to explore the second major finding why annual in- terpolation by FRB of its production indexes serves to put all, and even more than all, of the trend gap into years in which the FRB index is rising. This result is pervasive through time, showing up in the regressions for 1963-1974 and each of the three subperiods. This finding is tantamount to, but does not exactly support, a conclusion that the FRB-BEA gap is cyclical with respect to industry specific, though not necessarily general, business cycles. The gap does not appear to be explainable in terms of the various sources of FRB output measures or types of industries and their
380 PAPERS trend rates of growth. And it runs counter to a frequently alleged price in- dex inadequacy, the so-called list-transaction price problem. This prob- lem arises from the notion that the BES iS able to obtain only list prices of commodities and that actual (transaction) prices fluctuate procyclically about such list prices. If that were so, and the evidence, mainly that of Stigler and Kindahl (1970), is not conclusive, BEA deflation would overstate output during cyclical expansions, understate it during contrac- tions. Then one would expect BEA to fluctuate more, cyclically, than FRB rather than the other way around as seems to be the case. In looking at possible reasons for the cyclical pattern, it is interesting to note that in their study of quality adjustment in refrigerator prices, Triplett and McDonald (1977) find a cyclical component in quality change, which if not reflected in price indexes, would tend to reduce the amplitude of fluctuations in the output of refrigerators derived by defla- tion. The cyclical aspects of quality change as they might affect price and directly measured quantity indexes need to be explored further. Were it not for this finding, the likely place one would look in seeking to understand the cyclical differences in the two measures would be constant-dollar inventory change, which must be calculated in the BEA ap- proach. Such calculations must split the change in total inventories into its stage-of-fabrication components, because deflated shipments must be adjusted by the change in finished goods inventories and part, usually half, of the change in goods-in-process inventories. The latter tends to fluctuate procyclically. Finished goods inventories are generally assumed to fluctuate anticyclically initially, reflecting involuntary accumulation or liquidation, procyclically, subsequently, in the absence of perfect foresight. The difficulty of measuring inventory change is well known and has been and is being addressed in several research projects. More needs to be known about the possible cyclical biases in measuring inventory change by stage-of-fabrication. The third major finding of this analysis relates to the behavior of the two output measures during the 1973-1974 period of double-digit infla- tion, years for which the FRB indexes have not been constrained to bench- marks and the mix of monthly series physical quantity, kilowatt hours, and labor hours have been used to measure output. There are large dif- ferences between the two series during this period. The findings of cyclical differences are even more vivid during 1973-1974, perhaps related to high and generalized inflation, and there seems to be no correlation between the two series when the FRB index declines, only a constant gap in the direction pointing to symmetry in the cyclical relationship, i.e., when the FRB index falls, it does so by more than the BEA measure. In contrast, the data sources of the FRB measure seem to matter little. Again, these results appear to suggest that inventory-change measures be assessed.
Comparison of Industry Output Measures in Manufacturing 381 SUMMARY RECOMMENDATIONS The general recommendation arising from this study is that in seeking to reconcile FRB and BEA output measures, research should focus on the question of why the size and direction of the difference in behavior has a cyclical character. It might be asked why this strategy is recommended, rather than one designed to explain trend differences. The cyclical dif- ferences come through robustly in this analysis. The trend differences, though present, are smaller and potentially much more difficult to understand by direct analysis, given the many reconciliation items like weight and sac differences that would have to be considered at the outset. Furthermore, just as there is an interaction between trend and cyclical forces in an economy, so too, is there likely to be a similar link with respect to the measurement of economic variables. So discovering the reasons for the cyclical behavior of the FRB-BEA gap, may well provide in- sights about the reasons for the trend differences. Three specific recommendations follow from the general recommenda- tion. One appears to fall within the domain of each of three of the statistical organizations involved directly or indirectly in the calculation of the FRB and BEA output measures. 1. FRB should undertake to determine the potential sources of cyclical quality change bias and seek to test empirically for their existence where possible. 2. BUS should engage in a similar undertaking with respect to its price indexes, particularly those components of the producer and consumer price indexes that BEA uses for deflation of shipments and inventories. 3. BEA should undertake an assessment of the way in which it calculates inventory change in general and for purposes of deriving gross product originating by industry in particular. This assessment should focus on whether or not the methods employed and the data (other than price indexes) used in estimating inventory change by stage-of-fabrication contain the potential for imparting cyclical bias, and if so, attempt to quantify it. REFERENCES Gottsegen, Jack J., and Ziemer, Richard C. (1968) Comparison of federal reserve and OBE measures of real manufacturing output 1947-1964. Pages 225-347 in John W. Kendrick, ea., The Industrial Composition of Income and Product. Veil. 32 in Studies in Income and Wealth. New York: National Bureau of Economic Research. (Distributed by Columbia University Press.)
382 PAPERS Stigler, George J., and Kindahl, James K. (1970) The Behavior of Industrial Prices. New York: National Bureau of Economic Research. (Distributed by Columbia University Press.) Triplett, Jack E., and McDonald, Richard J. (1977) Assessing the quality error in output measures: the case of refrigerators. Review of Income and Wealth Series 23(2):137-156.
~ o o ,, == m I: is - or c) at r~ . - m ,`: ~ ~ ~_ ~ ~O ~ I _ O <O 0- m ~ ~_ ~ = E_ ~ . ~ Ct 0 ~ X , Ce ~ =4 US _ ~at _ ~4 5- ~ at Cal O ~ ° ~ O O . ~ _ 50 Ct ~ O ~ ~ ~ ~ O 5 I' itO- ~ LU o ~, m <¢ ~ S~ X f~V .Ce~ Z ~ ~ ^,_ . . . ~ O r~ _ ~ 1 . . . ~ ~ U: 1 ~ ~ 00 . . . ~ Lr) _ 1 ~ ~ _ . . . ~ ~ O 1 1 ~ ~ C:] . . . ~ a~ ~ 1 1 . . . ~ a~ _ 1 1 ~ ~ ~U~ ~ . . .. . ~ C~ ~ ~ 1 C~ _ ~ . . . O C~ 0- ~ O _ 1 1 ~ ~ ~ ~ U~ . . .. . . ~ O ~ _ ~ 1 ~ ~ O . . . ~ _ ~ . . . _ _ _ ~ . . . 1 00 ~ ~ . . . Oo ~ r~ 1 - ~ U~ 00 . . . ~ ~ O 1 1 ~ U~ O . . . 1 _ ~ C~ . . . r~ O ~ 1 1 1 ~ a~ u~ . . . O ~ ~ 1 _ ~ ~ . . . ~ U: O 1 U) U: O . . . _ O ~ 1 1 O ~ U~ . . . 1 1 ~ _ O ~ 1 1 U~ ~ ~ . . . ~ _. ~ ~ 1 .. . . . 00~0 ~ O ~ 0o ~ ~ .. .. . _O O ~ . . - 1 _ r~ ~ . . . Oo r~ ~ - - . . . ~ ~ _ 1 ~ a~o~ _ .. . 1 1 00 00 0 00 00~ ~ ~U: 00 ~ ~ ~ _ ~_ _~ ~ 0 ~ ~ ~C~ r~ _ 1 11 11 ~ u ~ ~ ~ oo ~ ~ ~ - ~ ~ o ~o - c ~ ~ - o ~ c~ ~ - 1 1 - ~ 1 u) o ~ O o 1 - ~ - 1 ~ o ~ r~ oo ~ ~ - 1 r~ o o 1 ~c) c} ~c.1 c) ~> s~5 - ~; - m <: ~m ~ ~m ~ ~m <: ~m ~m fi~ ~m ~ Lm~ ~ ~m ~ ~m oo ___ _ _ C~ ~5_ ~C)V) o ~=-50 ._ ~._ O ~X o ~D D 383
oft - -- 1 1 - so ~ - - oo ~ ~- - ~ of ~ a us or 1 1 1 - ~ ~ - o ~ r~ ~- - - oo - - - as .= OO ~ ~= ~ -0 a . ~ x ~ ILL 3 ~ =: V) ~ Cal ~ . . . _ ~ _ _ _ 1 1 on Cal ~ . . . ~0 ~ as ~ 0 . . . ~ rid O . . . O 0- ~ 1 rid oo us . . . Us ~ O . . . _ ~ _ _ _ 1 O rid . . . lo, a~ . . . ~ _ ~ - 1 - 1 ' 1 ~ ~ oO . . . ~ C~ ~ 1 ~ r~ u~ . . . ~ oO ~ _ _ 1 oO C ~ a~ . . . r~ ~oo _ ~ 1 ~ ~ c ~r~ ~ 0 .. . . . . oo C~ ~ ~ U~ 1 - - 1 . . . . 1 1 ~ U: OO U~ ~ . . .. . . _ ~ ~ O _ 1 1- ~ ' '1 1 ~ ~ ~ ~ oO . . .. . . ~ ~ ~oo ~ ro 11 1 1 - ~ c~ r~ ~ . . .. . . ~ ~ ~D ~ _ 1 1 1 ~ ~ O . . . ~ _ ~ C) C) 5- C) ~ ~ '~ o ._ X 5 O C c-t 3 x _ 5 ~ =-= O 5 O ~ C} S ~c! 5- 0 .~ ~ ~ C) 04 ~ ~ O ~ . . . C~ C~ _ 1 C~ ~ ~ . . . ~ O 00 ~ O . . . ~ ~ _ 1 1 1 ~ ~ _ . . . ~ V~ _ 1 . . . ~ 00 ~ 1 00 0 ~ . . . ~ ~ _ 1 00 ~ ~ . . . ~ C~ ~ . . . O ~ ~ 1 1 _ ~ ~ . . . 00 ~ O . . . ~ U) ~ . . . O O O ~ ~ Ir) . . . U) ~ C~ C) C) C) ~ ~ ~: ~ ~ fi C}C) C) s~ ~ ~ ·~ ~ ·~ r~ 384 C) :q ~ ~ ~ ~ - ~ CC C~ X . Ct ~ ~ ~ ~ .= ct 3 ct (L) s°- 0 ~ ~ C) Ct ;~
en ~ ~ a) an ~uO en 1 ~1 1 or ~ ~ or rat a) ~ or 1 or en 1 or on 1 1 a~ 0 0 a~ r~ ) - o ~ u) 1 1 ~ - ~ - ~ ~ 1 1 _ ~ r ~ oO - ~ ~ o ~ ~ o 1 1 cr ~ o ~ ~ o ~ ~ ~ 1 oo ~ u ~ u: ~ ~ ~ o ~ o o ~ r~ u~ oo ~o ~ ~ o ~o ~ - o - oo ~ ~ ~ ~ ~ - ~ ~ - 11 ~ ~ -~ u~ oo o~ ~ o - ~ c~ o ~ oo oo ~ - ~ ro O 1 ~ - ~ u~ ~ ~ - ~ oo o - - c~ oo ~ O r~ ~o c~ - - - (~- - o ~ ~ o 1 oo c~ ~ ~ ~ cr 1 1 ~ ~ o 1 ~ 1 : ) 1 ~ oo ~ o ~ ~ 1 1 u~ u) o 1 1 o o o ~ ~ o 1 ~ ro 1 ) ~c) ~;: ~s: c) ~ ~ fi~ m ~ m ~ m ~ ~ ~ ~ ~ m ~ ~ ~ _ C~ C ~ ~ "o W 385 r~ ~n ~ C) .~ 5 O ~ _ C~ Ct -
- z a~ - ro a~ - cr - - a~ - o a~ - oo a~ - ~D cr - ~D ~o - uO ~D - - . ~ ~ ~ oo 0 5' 3 _ c) Z a: ~ =O 04 Ce ~ C.) :^ i~ ~ ~ C) = 0 ~ ~ ~ O ~ ~ . . . _ ~ ~ 1 00 ~ u:a~ 0 a~ . . .. . . O ~ ~ _ ~ ~ 1 11 1 1 _ _ O~ . . . 1 1 1 oO ~ C~ . . . ~ ~ O 1 1 1 . . O _ ~o ll) a~c~ _ . . .. . . ~ ~ O~ ~ O 11 1 O oO oOC~ O ~ . . .. . . ~ oo _~0 ~ O 1 1 11 1 1 00 ~ ~DO O O . . .. . . 00 oO O~ ~ r~ 1 1 11 1 1 ~ oo ~oo ~ ~ . . .. . . _ O O~ ~ O 1 1 1 _ ~ ~ O U~ . . .. . . _ _ ~_ C~ ~ 1 11 ~ oO ~ . . . ~ _ O 1 1 1 . . . _ ~ ~ 1 1 . . . O O . . . oo O C~ - 1 O . . . ~ oo _ . . . O ~ ~ 1 1 ~ ~ ~O ~ ~ . . .. . . O ~ C~ _ _ _1 ~ - 1 1 1 _ ~ ~ . . . oO 1 ~ _ 1 1 ~ ~ U~ . . . _ oO ~ - ~ ~ oO . . . oo U~ ~ 1 1 1 ~ oo U ~O ~ ~ . . . . . . ~ ~ O ~ _ ~t 1 1 1 - ~ ~ ~ - ~- oo ~ r- ~r~ ~_ _ 0 ~ - r~ ~ ~ ~ O C~ u: _ c~ ~ C ~r~ ~ _ 1 0 ~ ~ a~ (` 1 ~ u~ ~ ~ ~ o _ _ 1 ~ oo ~ c~ - o 1 1 1 ~ ~ - 1 1 o ~ oo o~ ~ ~ r~ oo u~ ~ uO o oo ~ ~ ~ ~ . o ~ c~ 1 1 ~ ~- ~ ~ - ~ c 1 1 1 - o c~ ~ ~ ~ o 1 ~ ~ ~ - c ~ ~ ~ ~ o r~ - ~ u~ ~ -- ~ o o ~ c ~ ~ ~ o o 1 - 1 - 1 ,=\~ ~- ·r) _ ~ ~ ~ 0 ~ ~ ~ ) o ~ uO o ~ ~ - ~ ~ o 1 _ _ c} ~ ¢ ~: O: ~ · - ~ m ~ C) C) C) C) ~ ¢ ~ ~ ~ - ~ m ~ C) C} s: 5) 50 ~ ¢ ~P: ¢ ~ ~ ._ ~ ._ C) r~ ¢ ~ ~ ·- X ~;~ C} ~C~ C-' S°- ~ _t ( ~;- e ~t° .= ~e ~ $ ~_ ~ - >, _ 3 ct ~: _ ~ - ~V V) ~ 386 C~ . o . . . C~ _ _ 1 U~ . . . o _ o _ _ 1 ~ r~ . . . o ~ ~ - G) @ :~ L~ ·- U) r~ oo r~
oo= oo ~- to ~ ~ ~ . . .. . .. . .. . .~ . . . . .. . . . . . oo cut~ O ~ ~ ~- ~ ~ - as ~ ~ ~ ~ ~ c. ~ - ~1--1---- --~-C ~-- 11111 11 1 co is~ on ~ ~ ~co up ~ ~ to~ ~ oo ~ ~ to up to~ ~ ~- ~ ~- oo ~ - ~ to ~ ~ id -- - 1-- 1 - Cal 1_ _ 1 _ _ Cal ~000 Cal00~ ~Cal 00 ~ ~ oO cat rat 0 or ~ ~oO ~ 0 ~0 0oo0 oo ~ 0 ~r~ oO ~ oO ~O ~ ~00 - 1 c~ c ~- 1 - _ _ 1 - - 1 - ~ ~C) ~ ~ ~r ~ ·r: r ~ oo ~o t.~ ~ _ ~ L~ _ . . . . ~. . . . . . . . . . ~. . . . . . ~0 r ~ _ 0 - - c~ ~ ~c~4 - 0 ~ ~ 0 c~ c~ ~ 1 1 1 1 1 1 - 1 1 1 1 - 1 1 1 u: C ~ C~ ~0 r~ ~ ~ oO ~r~ ~ ~ ~ u ~u: ~ _ ~ 00 O O ~ ~ r~ O ~ ~ _ r~ - 0 0 0 0 ~ ~ u ~oo ~ - 1 1 - - 1 1 1 1 1 1 ,1 1 1 1 1 1 1 (> ~o c~ ~ - 't~ , (- ~ c') '~' r~ _ _ ~ ~ ~ _ ~ ~ Lr) r~ r~ u: ~ ~a~ - - 1 1 1 - - 1 1 1 1 1 - 1 _ ~ ~ r~ r ~ ~ oo r~ oo C7- oO a~ ~ 0 ~ r~ ~ ~ ~ . . . . . . . . . . . . . . . . ..... . .. ~0 ~c ~0 oo ~ ~ ~ 0 ~ ~ uO r ~O ~0 ~0 00 1 - 1 1 1 1 1 C ~- ~ r- ~ ~ °° . . . . . . . 1 - - O O ~ O ~ 1 ~ ~ ~u~ ~ ~ . . . ~ ~ ~ ~ ~ o ~ - c~ ~ ~ oo ~ ~ _ r~ 0 ~ 00 _ _ 1 ~ O ~r~ - ~ . . . ~ ~ o o ~ 1 C~ ~D - ~ oo o ~ ~ ~ - 1 c~ ~ ~ ~ ~ . . .. . . ~ ~ c~ ~ r~ 11 ,~r r~ c~ . . . . . ~ ~ ~ r~ 1 1 1 oo ~ ~ oo - . . . . . . - O oo ~r r~ _ - ~ 1 r~ . . . r~ ~ ~ - ~ oo . . . ~ ~ o 1 1 u~ ~ - . . . ) _ ~ 1 r~ ~ ~ ~ G~ . . ~ ~ . °~ °O O~ - r- - - ) c) c) ~c) c) :: ~s: ~ Q) ~G) ~G) O ~c ~.o ~1<: ~<1: ~6 ~6 ~CC 6 ~S ~6 ~- ~6 _ P:L~ ~- X~ ~- y CL1 '- ~ '- Cd ~ '- Ct~L) '- Cl ~.) '- 2t [~) '- ~ ~a ~ = ~ ~ = ~ =o ~ = ~ = ~ == _ ( ~,~, u' G.1 ~ ~ o O o _ - o - o _ _ V:O _ ~ _ - _ o C ~o - ' _ _ ~_ =i Cl, ·,) ~S , = ~ , ~C5 - 387
- - - o - at - oo - ~ ~ ~ o =. ~ . .. . . cr_ ~ us - -- - 's} O' 'A A' - on ~ ~ ~ cO ~ or ~ ~ ~ ~ ool 1 _ _ _ rat ~ ~ ~ ~ ~ on or ~or ~ oo ~ ~ ~on _ C~ 1 ~ - - - _ _ 1 _ c~ 1 o O ~D ~ ~_ c ~- - - - - - 1 - - ~ ~ 1 - ~ oo ~ oo ~oo ~ - ~ ~ ~- ~ ~ u: ~0 oo _ u~ r~ ~oo oo O -- - 1 1 1 1 1 1 1 - oo ~ ~ 0 u ~O _ cr ~ 0 ~cr o oo - ~ ~ ~oo ~ - ~ ~ - - ~ ~ 1 - 1 1 1 1 1 1 1 1 ~ ~ o ~ ~ ~oo ~ ~c~ ~ uo ~ ~ o o ~ ~ - ~ ~ ~o o o ~ o ~ 1 1 1 1 - - 00 ~ _ ~ ~ ~r~ ~ ~_ oo r~ ~ u~ o - o - - ~ ~ ~ - u) 1 1 1 1 1 1 1 1 1 1 -- ~ ~ o r~ ~ _ ) ~ ~ ~- ~ ~ 1 1 - ~ - ~- ~ ~ ~ oo~ uo o .. . .. . .. . ~ _~ r~ 0_ 0 0u: ~ r~ _ _ 1 _ _ ~ oo ~ 1 - ~ . ~ - o - ~ ~ - ~ ~ ~- ~ ~ ~ ~- ~ ~ ) ~ o ~ - - o ~ ~ a ~_ _ 1 - - 1 - - ~ - _ 1 1 o ~oo ~ ,~, ) - - - - 1 - - - - 1 - ~ 1 - ~ z c) ctS u) o ~ c) o ~ ~ - ~ z - 5) . - :^ ~ s~ t-, ~ o ) _4 ~ c) s: ; ~ ~: m ~ ~m ~ ~: m ~ ~m ~ ~m fi ~ ~ fi ~ ~ fi ~ ~ fi ~ ~ fi ~oo ~ o_ ~C~ .~ ' c `, ~, ' ~, a 388
or ~ us ~- - or~ ~ or~ ~ ~ ~ ~ ~ - O _ 0~ us 00~ 0 ~ ~ ~_ Us ~_ as _- 0 Hi 1 -- -- 1- 1-- 1 1- -- oo ~ ~ ~ 0 as ~ ~ 0 ~ ~ _ ~ ~o ~ . . . . . . . . . . . . . . . . . . . . . us ~ O ~ _ ~_ ~ ~ red ~ ~ ~ ~ 0 - 1 - - 1- 1 ~- - ) ~- - 1 us - - ~ ~us ~ ~ ~ or ~ ~ do us - ~or . . . . . . . . . . . . . . .. . . . . . ) or us do ~ ~ or ~ ~- - - - ~ - 1 ~ ~ 1 =) _ ~_ 0 _ ~ 0 ~r~ ~ ~ ~ ~ ~ ~ ~ . . . . . . . . . . . . . . . . . . . . . ~ cr _ ~ ~ ~ ~ ~_ _ 0 ~ ~ _ ~ ~ ~cr oo 1 1 1 1 1 1 1 1 1 1 1 ~ ~ ~ ~ ~a~ oo ~ 0 oo u~ u ~_ u~ ~ ~ . . . . . . . . . . . . . . . . . . . ~ ~ - - ~ ~_ ~ ~n ~ ~n _ ~ ~ 0 _ ~ ~r~ ~ 1 1 1 1 1 1 1 1 1 1 1 1 1 1 _ 0 _ r~ ~ _ a~ ~o ro ~ ~ ~ ~ U) O 00 ~ C~ ~ . . . . . . . . . . . . . . . . .. . . a~ ~ 0 ~ ~ 0 u~ ~ ~ 00 0 0 ~ _ ~ ~ ~ ~ O 1 1 ~ ~ r ~0 ~ ~oo - ~cr~ ~ ~ u~ ~ C~ _ oo oO O - O O oO ~ O ~ ~ 0 ~ 0 ~ r~ oO~ ~ _ ~1 - 1 1 oo ~ ~ r~ ~ oo ~ u~ ~_ ~ ~ ~ ~ ~ . . .. . .. . .. . .. . .. . .. ~ ~ ~ ~cr 0 0~ ~ 0~ ~ __ ~ ~ =\ ~ ~; - 1 1 1 1 1 - c) c) c) ~ ~c) c) c) ~s:: ~`: c) ~ c) t> 5 - s~ s ~s ~s" ~ ~ ~ ~ a ~ ~ ~ ~ ~ ~ ~ ~ a ~ U ~U) U~ C~ ._ ._ ~ (, S" ~ S" ~_ ~. ~ ~S ~ _ c~ ~o ~ 3 389 o 1
ran an - - o - a~ as - oo ID an - u~ ID - - is .= as ED red - a Z; C) en cell O 5 _ O ~ 3 _ c) Z - ._ ¢- ED 3 0 ~ V) _ _ oo ~ Ored O ~ . . .. . . _ 1 1- - 1 ' '1 _ ~ ~ ~ oo~ U) ~ . . .. . .. . . _ ID~ ~ -O ~ - -1 -1 - - 1 1 1 1 1 ~ ~ 0~ ~ cad~ ~ ~ can _ ~ an ~ . . . . . . . . . . . . . . . _ O ~ ~ O ~ ~ O o - 1 - - 1 - - _ 0 ~ r ~O ~ ~° . . . . . . . . . . . . . . . _ ~_ ~ ~ _ ~0 r~ ~- 11 ~c~1 __1 __1 - ~L~ 00 ~ 00 ~00 ~ ~00 ~ - . . . . . . . . . . . . . . . oO ~ _ ~ ~ ~ _ ~ ~ U ~O O O - 1 - 1 1 1 1 oo ~ ~ O ~ ~ ~0 ~ ~ C~ . . . . . . . . . . . . . . . O ~ _ ~ 0 ~ _ ~ r~ 0 - ~ ~ ~- 1 - ~1 1 1 1 1 1 1 ~ C~ ~m. ~ ~ ~ ~ . . . . . . . . . ~ _ ~ ~ ~ O - 1 1 1 1 -- o - - (~ 'c) O' . . . . . . ~ oo ~o o o 1 1 ~ o ~o o . . .. . Oo r~ _~D ~ ~ ~ r~ . . .. . . ~ ~ o ~ ~ o 1 oo r~ ~_ ~ _~ u r~u~ ~ ~ u~ ~ . . .. . .. . .. . .. . . - ~ o~ o ~ - ~o ~ ~ o - _ __ C~ 11 1- 1 111 1 _ ~c~ ~ a~ 0 ~ ~ c~ . . .. . .. . .. . .. . . - ~ Or~ ~ O~ O O°N ~ ~ ~ ° 1 1 1C~ ~ 1~ -- 11 =. ~ ~ ~ ~ o ~ - -~ - ~ ~ ~ - ~o ~ -~ ~ o~ ~ ooo o~ o _ _- -1 ~D O~ ~ . . .. . ~ ~ ~- ~ 11 1 ~t ~ ~_ ~ _~ r~ u~ . . .. . .. . . 00 0 ~oo _ r~uO u) 0 - ~ 1~ 11 c)c} c)c)c) 5)Q) ~s~ ~ ~ ~m ~ ~ ~ ~ ~ ~$ <: ~ ~ ~ ._~ ~ ._ ~ ~ ._~ ~ ._~ ~ ._ ~ ~a~ ~a ~ m~ m~ cc~ ~0 _ U~o ._ _ ;: C'' 0 ~._ ~ 0 ~ C~ S £ ~ £ ~ X ~ &, ~ ~ 00 o =' ~c, 3 ~ `,, o 390