A STATISTICAL APPROACH TO GO-C RATI3 t-POITICS by-Howard Rosenthal K Center for International Studies Cambridge, Massachusetts vlay, 1962 *1 In 0 EFRATA Location /p. 2, see, 1.11 Corrocted Incorrect 04., x < z< 1.13 193 t p. 3, line 12 p 5, line 18 V v V O V section 1.2 to section 2.2 ) p. 6, line 4i (see see p. 6, line 16 of w and e p.8 p. 13 the equation for the error function should contain 2 ~ 9 21 2 - 721 P(721) .2 6.97 P(6.97 Table II A the growth rate for Brazil is "L" Table II B Group II and Group aI Group II anI Gro' Figure I z alogr x (see see. 3.3) of wande logy I A STATISTICAL APPROACH TO COMPARATIVVE POLITICS by Howard Rosnthal . ;roduction 1.1 Among the key economic, d" nics nd c ahic o r oad markero of socitl 6oIcncer, there eXits ae are ttl membars have for all practical purpoo, bo attained. Exclud- ing aSe elass includes groea nat f inite lnal protductv p013C hil adios st t 0o1lq as literacy o 1 The vti Onergy, newspaper copies (por capitz, etc. c.ubd the distributions of these variables amngnr nations and a ong of . to th uppr boun .o . n i.Unation ib ono topic of' thi.s Ppoe ariablev The basic dJstributioa that characterizes theoe thet t-normal di it of which ribution (see 1.)) ubsets yield and moder On the inCternati riomal- pIar t; of log traditioal., an empirica. classfication of the natiorns.' i1l be reflected in the three d Ughon he oghTh" lo A i4 , rnitic nat;', Another break down would be betweenu bo stens ttrowth no-tiofle aI na -WiU2' and mature notin pattern, oi p CAizt h tion parmters apirica1y, the, variables appear to be 6 itribuIted mong subsets of the nations in log-normal fashion. the paper portrays a grt;Vh proces under which a 1 would result cubset is necessary to divide the population into thee s choracterized by a deif'Qrent se f mn the ntC The vecond sectic og-noma d i st Ii.u In the third section, a body of empirical dta is p See Daniel Lerner, The Passn of TrditionaI Sociytv Gnicoe Aconept ivtive fromnn tht oft). Ro sto h stage Gtrowth, London (1960o 1 C; (IV9 8 ___-V dL W 9 discussed uhile the fourth diccusse3 applications of tho findigs. broken o ba h.. graphic-l techniques have been applied. nly To aato According autom-ated analysis can be developed in the com.ng cunmer. only a liaited range of possible apolications is presented. 1.11 If for a variable 0 og zz , is normally d istribded tle say that x is log-normally d istribixted. 1.12 If z o log y is log-normaly distributed, then - normally distributed from vell-mown properties of the norm' This implies that if a variable, say persons/vehicles is importAfnce. C r e pp10ies In the physical sciences, thee-nor lo -noml also log-normal. then its reciprocal, say vehicles/ca pita is 1J3 n ditr io Thore is a suggestion of an analogous phenoienon in the socia. sciences as the investment rate f:Qr tie;alh or birth rate for popui.ation oealth tends to be a reasonably constant percentoage of the existing population. 1.2 grooth' "Mitles Among the possible applications staisa the uise od tics in making inferences about " political deve.opmen stati The log transformation has facilitated an analysis that the author beievos superior to earlier work based on simple avreraging. 14Te The objectiva of section 4 wil be to show that indices cannot be simply "added up" to gi.7e an "avrage' picture of developmert, On any given Variable le position relative to the tir.ld C 3 Yh tics (T 4 the argument runs, the unctions of the variable yin expectation.-ereatIng and eut must be considered. Instead - aWLd2txJ. log-normaistribution is discussed by G. ill on. comnute the nation G Then.rse ighti1ng Hordan, S Lxnon (1960), pp. 81-106, and Type--Token pp. 425 and passim. . Particle Statis ahematica, 8-aenha5 See the "Concluion" by Jamen-- oloeman in The 1olitics of the De-velopinlpU and T:920 do. Princeton Areas, Gabriel Almond and James Coleman, =,,vrtt Hagen, "A General Framawork for Arialyiridg Economic and 41olitical Change" 9 Center for International Studies,.'timeoo, Cambridge (1961). a.xpctation-satisfyin~z variables positively and expectation--creating va abl.e negatiel, be a dovelopment or instabilitr nerato' d Thic indox yields a U-shLe~d rolationsh p vitii a batbory 'of indices of potUial d 2. 2,1 S-omo likely conditions for lop:-normality* The settlement of an area, the introduction of automobilos or rd:Los to etc. can be thought of av dating from someO that area fctive "otarting time" (the "starting time" will acquire a specific Assuna bela) t f(t) t Tha n fA (t .meaiIn c i0athema in normally dIstributed smzh sartiny time t T- iu 0 t 2 is normal such lthat -to 2 Where t in application, t can be taken as the ti ne 'ofmeasurement6 2*2 lop enat Assumo for some variable (of the "imitlesa is logarithmic with time; log ya growth" class), growth rcisel N t to co'. y Further assume that k is a constant over the population. Then it follows that "S deviation normal Trlth mean x kWand standard X SiLncc 20s normal, by definition~y)is log-norrnal.* t 'U The assumption of a contant growth-parameter, k Interpretation: is not as"far £ront reality as might be inagined2 0 In generals the deviation of log k sceeris to ba substantially less tnan the deviation of 10 y (see 3.3). (ry" refers to the growth variaol e.g. GNP/capita. Such a condition would imply that is the important variable* 2 .3 A Pieceuise Linear Model Let us assume that x. the log of the growth variable y, is ro3.atod to x,, by three different constants that apply over different ranges of That is, lot X 1 2 2C k3 + c '+ c2 - X--- x 1 +cx X2 x2 Let us further assume that'Y is normal as In 1.1 Then it folloxs that the distribution of x is the composite of three different norma distributions, one of which holds for each k. Presented in 'tabular fori, the results may be given as: Range of x :k 2; '' Normal distribution parameters Mean S.D. al. k~ C3 '3 j; ki X2 2'0', X x k a A graphic illustration of the behavior of x withand the rsulting density function is presented in figs. 1 and 2. Emirically, a three-piece "piecewise log-normal" distribution fits much of our data. While no assumption has been mDads about the values of the k's, the k 2 implied by the data is generally larger than k, and k3 This accords with a traditional., transitional, nodern nodl with the two breakpoints signifying takeoff" and "maturity. Of course, any distribution may be approxii#ted throuh a 'ee of normal segments, and any distribution may bp seg mented to appro::imate another. The roason for using the normal is that it has a reasaonable relation to a growth model. While each segment does call for three additional pardkmeters, c, k, and x, these may be taken into account when pnrforming tests for goodness of fito 2.4 Implications of a b variate log-normal distribution of a gvro- th variable and growth rate. To conclude this section on conditions under which log-mormality can occur, we should like to examine the joint distribution of the growth variable cx,and the growth parameter k, from section 2.2 and see if there is a reasonable imnlication about the "starting tiges", to. In this secbion, V will be a constant within any nation variable over the set of all nations. and a random Givan two times, tl and t2 and corresponding y, and y,, k shbu. be r iven by log y - 2.o7 y2 i Thus, n o stirmated from empirical daca, o:d it o tprs to approach lo-norial also oxnra 2 t. accord with oit Lbutions -,a) orm. (so cc, in th ln:- 3 Theso lo noldge of the skewd (aCnd distribution of irco.mr, mobility,(growth), c within r.Aional populations. If k is lot-normal, then if we'define w log k, w is normeal Lo joint distribution of w and x is bivariate normal, us assume th with correlation peritted, for those nations with large x may tend to have had a large growth constant as well as an "early" -to Let us normalize x sich that its marginal distribution is unitnormal. Then, the joint distribution may be expressed in the bi- variate normal formp where ad i 4' of w and 1 is It arie the mean and s.d. of the raarginal distribution the c variance0 is well-known that the conditional distribution of x given w of the form g ) x -g W". This distribution leads to inferences about the starting times et us calculate Since x k(t - o t) by assumption and w.l ey or since -t 0 I, x 10 Therefor lo' is a simple lindar transformation of iy , the starting times given the growth constants are normal with the standard deviation proportional to 10 1/k and mean proprtionalto (l/k)(log k -mlog k)0 Thus, there is the rather reasonable result that for a given growth constant growth will "beyin" as a normal random process 0 The behaxior' of the parameters with k is also reasonable; infinite growth must take place in a compressed time interval; no rowth has an indeterminate distribution 3.0 Examples of log-normal variables Efforts to find a theoretical basis for the occurrence of a log-, normal distribution should not dominate our empirical results which clearly show the log-normal character of growth variables. Both graphical andstatistical methods are available for examining data for log-normality. This section will begin with the latter, which has illustrative value, and conclude with the more rigorous statistical testing, -SBA 3.1 A preliminary examination for log-normality may be made with the aid of log-probability graph paoer, a varilant of the more familiar* probability paper. On log-probability -ptper, the vertical axis is logarithmically spaced. The horizontal axis is spaced according to the unit normal cumulative or error function, given by W(x)= (2T) e dy It follows that, if we plot the cumulative percentage up to a certain va.1e of the variable against that value of the variable of log-p7.hbability paper and obtain a straight line, then the variable Is lo:normally distributed. The value for cumulative percentage equal to .50 rives an estimate of the mean and the value for cumulative perciz: -agec 10 and 84 give an estimate-of the range coveredby-The standard deviation. If convenient sub-samples are taken (with N100 for examplo), no percentaging or other .computation .is necessary, and a very rapid check may be made as a preiminary to the chi-square test discussed in As an illustration, the first 100 basic political units in an alphabetical list were picked and their population density per square mile for 1950 was taken as the test variable. In figure 3, we can see close conformity to a log-normal with mean at log 61 persons/sq l miLo A basic political unit is defined as either a nation or a colony in I IIs ., the standard deviation extends to log 13 persons/sq. mic and log 290 persons/sq. miD A linear plot is also obtained for a wide r ange of growth variables over the set of "developing" countries. We have used in these cases the data provided in the appendix to The Politics of the Developing Areas. Although we ill eventually want to include all units, a compact and reliable source of data had initial advantagos. As figures 4-6 disclose, the log-normal distribution holds reasonably for persons/radio, persons/vehicle, GKP/capita, and persons/docto:, and daily newspaper copies/capitao a 1956 unOo (For the newspapers, we have used source that also contained data on the developed nationso) We havi also found a linear fit for persons/telphone and energy. capit ne smll number of countries nvoed (aproximat y 60 in each case) is offset by the generalty-T the distribution. In figure 7, the line for each variable in figures 4-6 is drawn as if all variables had a common mean in order to allow the reader to compare the similarities In standard deviations0 The deviations for all variables except O P per capita lie in a narrow range of o4 to a90 (in log units)o While this coincidence may, like a high correlation, reflect the systemic character of development, we have been unable to develop any firm interpretation* The fact that a single set of log-normal parameters serves to describe worlpopulation densities or growth variables in developing nations doesno ply any supra-generalityo 1. Almond and Coleman, op. cit. In the cases where a -0 single set suffices, it appears that the units must undergo similar growth processes, Where, as was mentioned in the introduction sub- sistence or saturation levels occur, a single set should not sufficer Our newsoaper circulation data clearly show a saturation phenomaenon when the developed nations are included* For those 25 (mostly developed) nations that have large newspaper circulations, the lopnormality that obtained with developing units no longer holds, The curve of figure 6, with a constantly decreasing slope shows the development of saturation0 A subsistence bottom is demonstrated by the data on real per capita GNP for 1961 as presented by Rodan: While a strict linear plot is obtained above $120 per capita in figure 8, the lowest 15 to 20% of the nations appear to bottom out. Rodan has included several remote or specialized areas that Almond and Coleman omitted (Bhutan, Muscat and Oman, etc.) where the extent of poverty is kept in complete traditional balance. Just past the $120 marker lie those nations with the bevinnings of industrialization and/or export agriculture (Belgian Congo, Nigeria). We are led to the sugnestion, with reference to the theory of section'l, that the breakpoint on the curve emoirically distinguishes the "traditional" from the "transitional." Books and paper variables, as shotn in figures 9 and 10, also exhibit breakpoints although the data is particularly incomplete and unreliable. In the case of booka, we have a linear plot for -the first 50% of the nations up to 21 bookr/iO0,OO0 capita. After this point, only l. P.N. Rosenstein-Rodan, International Aid for Underdeveloped Countries o, Ca ridge,9. Center for International Stuies $urooean -units are included and saturation sets in. With paper production, after 50%is passed, an increased slope occurs (Uakeoff?) followed by a shar-o saturation with the exception of the United Stateso With newnrint, a single set of log-normal parameters gives a rough We notice that both the deviation and mean are maintained over fit. an 11 year period, the pre-orld War II 3.1.1. This p rhap eflectS a aen ral read4justment to levels with some shuffling of position. To conclude our graphical illustratione of log-normal behavior, we have two examples using political units within a single nation, the Unita Stawcso In the automobile example of fi-ure 11, there is a clear' saturaztion breakpoint brought about by the depressed levols of the Southern states * A tentative analogy can be drawn betweon the underdzvoloped character of the South (especially in 19409) and the undedxveloped nations and their corrosnonding log-normal properties relative to the developed s tates and nations0 Our second example returns to population density per square Figure 12 shows the distribution over the 48 continental states botween 1810 and 19h0. InitiallyM.Linar pieces must be used to describe the data, the upper piece for the settled Eastern Seaboard and the lower piece for the developine interior. progrresses and as growth becomes more uniform, As time the deviation of the lower pie.ce approaches that of the upper until by 1940 they are nearly identical. Here is an excellent example of how presentation of the logunormal distribution can illustrate a developmental process. A few highly urbanized states fall below the breakpoint0 This perhaps reflects the greater availability of public transportationo There are two further points of interesto One is that the point of intersection betwoen the two segments occurs at a higher level as time orogresses. The other is that the deviation of the upper softwnt is naintained constant although its overall growth rate fluctuates. Clearly, these facts are at variance with the piece-wise linear model of 2.3. ted They suggest that it will have to be soobistica to allow for an increasing saturation point as technology progresscs. Additionaly there is a suggestion tIiat a type of stabilization occurs There the units maintain nearly constant ratios between each r the overall growth rate, other ate of ra i e absorptive capacities Thse This implies stablization problems of r uildirng should not, however, distract us from our'irain task, the presentaticn of empirical evidence on the log-normal distribution of growth variables* 3.2. A more rigorous indication of the presence of log-normality than graphical methods would be the successful application of a chisquare test for goodness of fit. Our graphic examination has indicated that, of the developing units in Rodan s data, those with greater than $120 real G0JP per capita should form a set over which real GNP per capita is loc!-normally distributedo There are 90 nations in this set for which we have estimated the mean and standard deviation of thelog at 2.37 and .25 respectively, An eight-class test gives the observed and expected frequencies contained in Table I; The test satisfies the 20% level. .M13 . -TAlE I I III .15 IV V VI 12 -10.7' 9 10 10.7 12.6 9.5 12.2 VrI 5 VIII it Degroos o.f freedom 91 o-2.11 2.11- 2.21 231 -2.21- .- 12.2 :5 12o6 10 . .15 -- 3 0 0 238 - 2,44 2. - -7 7 - -141 2312.238 - - .7 2.64 2,54 2,64 -1 8 el - i1 7 5 2 o20 o(70") PfR) del Rane mransformed to unit normal Range Expected Observed Class The combined graphical and statistical results clearly warrant cont ineed intereet by social scientists in the logp;normal distributiono - - --- - - - - ,------ ---- --------------- -------- -- 7 - - - he3.3 In our worc with American population figures we were also able to compute a value for k based on the 1890 and 94031 urea. In both of these years, one set of lop-normal carameters described the distribution of densities over nearly the entire set of states. be seen in figure 12 . This k also plotted in linear fashion as can Its deviation differs f rom that of the population itself by a factor of 2. The result blends with the investigations of section 2*4* Aplicat-ions to po0litical scienceo Tho discovery of the generality of the log-normal distribution offors some inindiate advantages in comparative studies. Any log-normally distributed variable may he normalized with respact to the mean and standard deviation. A d eveloping nation's position relative to other nations is given by its normalized value which may be comp red to normalized values on a series of other variables. Thus, a nation s average position on the international scale may be computed as well as the variation (balance) in position. While many results may not be sensitive to the method employed, use of normalized scores has a clear geri6ral preforeic agirg rank orderings and other techniques that have been applied in the past* As an illustration of the use of normalized scores, a computation has been developed to test an hypothesis on the nature of political development As mentioned in the introduction, Hagen and Almond and Coleman have computed some sort of average position on a number of indicos\(1) and have attempted to show a form of linear correlation between this position and "competitiveness" of the political system a 14 0 , The "competitiveness" concept, in addition to its subjective difficulties, has the weakness of beinp unstable. In the year that elapsed between the Almond and Coleman book and the Hagen paper, Hagen chose to change the classification l4oreover, "competitiveness" in practice seems to of 5 out of 60 nations, place too much emphasis on formalstructure and too little on the crucial outputs of the system0 An alternative view of the developing nations would emphasize the t ransition from one type of legitimated and perpetuaing power structure to another with an intervening crucial period often tormed "Ithe revolution"o The events in the revoluti.onary period tend to acquire a defining reference for future decisionsQ (The classic example is, of course, the Soviet Union.) Ve 'Toul33A focus primary interost on the conditions under which the "revolution" occurs takes. and on what form it As preliminary indices of "revolutionary" eventa, we will take leftist takeovers, expropriation of Western property, and internal wars, The events in question should occur, in the roughest terms, when the creation of expectations outruns the satisfaction ofexpectations. Some develop- mental variables such as income and medical care tend to satisfy expectations, Others, such as mass media and transportati.on, we would argue, tend to build,. more expectations than they satisfy. It follows that, ihiattempting to predict the unstable, "revolution" prone subset of nations, some indices must be weighted negatiyely. In terms of the normalized scores we have experimented with an instability index, I, such that I - 2x GNP/capita + Doctors/capita - Radios/capita -Vehicles/capita We would epect the unstable nations to lie in the middle of the rangec of one end will lie the familiar examples of "good' deivelooment (India, the other extreme will li Turkey). At At those nations whose (primarily income) levels of | development have not arrived at revolutionary ootential. Tnble 2A presents scores for those units for whom Almond and Coleman have provided a complete. set of data0 Splitting the ordered vst in thirds in tablo 2B e ind ih osctod association (no lon er a linear correlation) with the three aforemention'd indices of "revolutionary" eventsi While small numbers were involved, -o did find a preference to straight addition of indices. This is mainly a result of the index s classification of some low income countries (India and Pakistan Cre examoles) into the first class and some high income countries (Cuba) into the middle group0 As a byproduct, the instability index gives an association to ates of economic growth that is superior to using either real GNP levels as the predictor or the Almond and Coleman indexe. These results are presented in Table 2 Cc 11 the instability index appears to be '-ulled from a hat" it is so than the "add 'em all up equally" indices0 no more In fact, there is perhaps more logic to weighting income double than to leaving it equal to the others. What we wish to offer, in any case, is not that the present research offers any solid proof but that it extends the point that some forms of growth are clearly preferred and that growth on any one dimension does not always make a positive contribution to a nation's stability. The succesu1-tiiMh growth rate, low violence rate nation perhaps must e-tz its cargo before the media cult arriveso 1 The third class alsO includes aznumbez. of "settler" colonies in which media and vehicle consumption "ave been atypically high, TABILE IIA NATIONAL INSTABILITY AND GROWTH DA.TA (Note: The classifications made are ex-tr ,emely tentative. Although th3 m thooogy must eventually be justified, no attempt will be made in this preliminary "udyo) Nation bility Index" Leftizrt Takoover LargeScale 1945-62 Firopriation 1945-62 Lo v (Inrnal row t h Ra ars/ 2opoo,cOO Capita) 1946-59 Venezuela Israel 147 15o2. 13.0 11.1 10.6 El Salvador 9.3 1,8. 0 2,1 1,9 1.3 Columbia India 9.2 9.0 0,14 irgentina, 8.6' 7.9 7.9, 6.5 6o5. Uruguay Nicaragua Ecuador Lebanon Pakistan Burma Nyasaland Phillipines Thailand Turkey Congo(Loopoldville), Sudan Viet-Nam Panama Guatemala Dominican Republic Costa Rica Honduras Iraq Cuba Brazil Mozambique Malaya Tanganfka Union of South Africa Paraquay Iran Indonesia -U.A.R. Libya '2 5,3 5.0 4As9 4.8, 14.8 4-. + 3.9 3.7 3.7 3,0 2.8 2.6 2.3 .2o2 + + 7 2.0 1.6 1.6 ? 01. ++. + 0o4 2.5 03 1.0 nodo 0 .9 O7 0-6 L L S L H H -11 H L 1.1 n,,d4 2.7 2o1 1.0 S 2,5 I S 1.8 105 2.2 048' n.d. 1..1 0.1 105 2.1 1o4 0.5 143 1.9 GROUP L S L L L GROUP L nLd L L-S L4* TABLE IIA (Continued) Nation I ("Ins a Growth LargeLog Rate (Interna Scale 1945-62 Expropriaton Wars/ i0,C00,000 Capita) 1946-59 Leftist bility Index") Takeover 0 Ceylon Ghana Nigeria So. Rhodesia Jordan Peru Algeria Mexico Ethiopia French Equitorial Africa Haiti Chile French 'West Africa Tunisia Kenya Angola Morocco Bolivia Key: -1 7 -2.0 0. L n.d. -h~h 1.6 -4.6 0.9 -5.5 100 L 028 n0.6 L GROUP L III -4 4 H L -548 250O 1 3 n,d 1.3 -8o2 -8.2- 1,14 -10.1 0.k 1.3 1,2 -10 _7 -14.6 +7? L L L L L 2.0 Countries include all with full set of GNPp radio, doctor, and vehicle data per Almond and Coleman, opo cit. nod.: Sources cited had no report on these urits, No data, H,LS,: High, Low, and Stationary growth rate.. per Rodan, op. cita * Uruouay had no incidences of internal war, The arbitrary score of 0 places it lowest in rank order. Egypt is low, Syria stationary0 -Source: Harry Eckstein "Internal Wars: The Problem of Anticipation," *N* A rep.,rt for the Smithsonian Institution, mimeo., Washington (196?), We have used the total incidence of Eckstein's "Unequivocal Cases." This category includes "Warfare, Turmoil, Rioting, Terrorism,ating, and Coup." ? These symbols are to indicate the degree of appropriateness of the classifications 1 0 TABLE1 1I GROUP DIFFERENCES IN STABILITY Takeover All +, * 1.27 0 Group I Group II oroup III * Expropriation Internal War Means 7 2 6 1,5 1 l2 +?, and ? cases from Table IIA have boen c6unted A t-test of the hypothesis that the difference between the means of Group II and Groupts I and III is positive was successful at the .01 levela Cases with no data excluded, TA BLE IIC COMPARISON OF GROWTH RATES AFD INDEX CLASS Growth Group I index (4 basic indices) II III H L S 6 0 2 10 1i 1 2 -32 3, Almond and Coleman (11 indices) 5.b 2 Growth Group S8 17 18 Growth roup * Groups formed from rank-order Reai 3 III 3 * 2 III Capita H{ 1 L H S 10 5 9, 23 20 3 L 12 11 S / j-c) a. 5 For the day after The imimdiate task is to buttreess where pocsible, our data on undordovelooed countries and further toot the hypothesio that creation va satisfaction is crucial to political stability would include varying, the parameter tre~nd dat tur to in of si These tests of the instability index and lookins at With trend data, we could look more closely at s-ubsience an k fcts as variants of the ." eeral lor-normal case The Orwth and the ability, to c tch up to developed nations bi Ihou be lcr importance as n6sition relative to other developing nations 9d data on developin d penage of comuter to expctat c nations c an be explored quickly w th an rograms The sv cz ing step will be to turn eetern Eurone and the UnitedZtatos where detailed lon av~alale on socio-economic variablez. term data is After trying to estinate the rela'ic of those variables to expectations in n particular time period and us n tr dat or parameter estimation, we might try to predict which sociological strata (locality, class, etc*) are in a state of high expectation frustration and to test this prediction by an appropriate survey* hio V. ... L- s 0 Relstinsy bcit,,cu und :F 6c c1~ 1! ~AiIAkX '~1 ci~ (p t-to I/ 0 cj) FJGRnE 2 a /, / MltttiollO o12'sIty 11 10 functioa, for tbre-Yiece mdel FIGCURE3 Log-normal cumulative distribution of prsona/dqod "basic political units,"i 19509 . in 10,000 Persons Sq. Di. 1.,000 CUNHiATIVE PERCENTAGE indicates approximate location of actual data Positions of some units indicated for illustration. . IGURE1. Log-normal cumulative distribution of 13ersons/doctor, persons/radio, persona/veh:: iOOCO Source; Almond and Coleman 10, Doctors x-65 .adios Ns26 Vehicles N-62. Persons Unit 01 50 CUULATIVE PERCENTAGE o Points from doctors data Points from vehicles data ( Points from radios data 99099 ~pi~: hg~o~iiicurhftive diirytribution 6 'Sou rceo Aiyaord' and~ C o lema n NU~L~~EE PERCETRIAC:E 1000 3.00 it IF A FiGruRr, 6 ) I Sourco: V1NESCO copicS. - 10 O0-01 :16 314 M-1TAGE.? GtJIUTATIVE PER MC 99099 \01 EoTwoouI asded9aqq !L7aam po7,-, Vinourx a indmo0 sadOT13 Tv=,Ou-'Oq U Figue 8:RealG/Cpt Source: Rodan -. 100 4 yd:us I., Virgin I., Grenland, Canal gone arbitrarily excluded rill "'6'.,. Co a * S I. ii A-'0 (9. H ""- ~ ~5 V.' ~ £ I j 'I I - * 4 (21 Cd S. 4. .0 12) K) 4-; 0 C; C> p Jo '-A - H 0 ~ t*~tn '-3 c:~ f~3u '1-~ - ,% 44 4.,' C' C 4 4 I '4-J b -- 'I 4% ~, 0 fl "'''.4 lx t P 4 ) "'"6 k. ('03 4 I r~ C C 44 (3 C "5 ;..i *'tJ ~2 6?) Ul.. K9r t 'ti I $ 3 .4 4 I '2 k 4 ~4fl4 -I $ Figurc jr A Cumulative distributions for newaprint and papel Cpnuption ic1Sor: TJESCO 1000 Kg/head 1939 1950 100 2. Cu .0 ATIVE PEFENTAGES Auomobile regictration/capitS 19h0. Source: Sttistic2 - Abstract 'of the UOBS ;99 CUIt7LATIVR PERCENTAGES ( ?) ba FIGUREf 12 V Cutiu2ative di.stribuuiono of poptilection density in thL3ic~ pr t CAP,- Uiltod Stcetas Source: Sttfistical ID2flsity Poqson 4 lo 64 '/sq, 0001 - 16 - S ".0 deviations-of' actua]. data .990)99 rnio * p * First draft-not for circula tion data to the graphical e:ination Txhis appendix adds statitical Distribution parametcr! ar, presentcd alon- 'ith th in thu text.; 0 for chi-souare tests for goodness oCf ri Comrtatio; ~' Rodan data Variable p abovo 90 countries: iea 1I20 real. GN/capita~ njti -log oficn ""*.L75 6329 h GNP/capita ln(GNP/capita) 0178o7 695 Perons/doctor ln(persons/doctor) 8682o, 9o06 5.16 5462 5o571 Parsons/radio ln(Persons/radio) 371,4 4173 322h.4 Persons/newspaper copy 4.4?1 ln(Persons/newspaper copy) 2 2. 6o97 8 This work wlas done in part at the All data per capita OtI Ch -Squarc 137o8 M66 7 32 25238~ 8600, 10 66 coutries AnA sean m Lvel atisfied *080 1o29 137o8 313o6 Persois/telephone In(Persone/telephone) 322.9 0,66 0,.,56 Almnond and Coleman data: Mean S. D4 330.7 5.59 in(persons/vehicle) t) b 10 4235,,5 in t in th oss (not the log r The oigt-class chiquE discumd in tne text was usod in alcs. na2tur.l lor arithm dv 174 1010 626,3 264 o02 1.32 506.3 13 85 lo7 18 8 3 819,7 io,6o io0 83 .I.T. Corputation Center 00 pE 0 & Almond and Coleman data: Variable MIean 66 countries (continued). Anti-log S.D. Chi-Souaroe Level Saet 13fie d onergfl3y convo/capita ln(onergy/capita) 4 a O32 -1.87 OO45 45 1o22 15o03 01 Interpratation of Rlsults While the chi. esquare results for veihicles and dgatprs are extremely satisfying and the results for income for the remaining variables. atisfac"ory, there is a poor result Whether thais revults from bias in reportin communications and production data, saturation effects at the t ails of the distribution, or theoretical deficiencies is a topic for future invuestiga ion In any casei there seers a clear oreference to using log values in place o0 the untransf ormed values. The standard deviations of the untransforrod variables range over nerative values of persons/doctor, telephone etc. Cnd thus reflect the high y skewed nature of the basic distribution. -~ jaare r~ther reinforced in our argument for the log-normal by correlation matrices oresented below. That the logs hne ive substantial and uniform increases in corielation between variables suggests that the variables are pot-er funct.ions of one another. the exairination of scattergrams. This has been tentatively confirmed b5 . CORRELATION MATRICES FOR PER CAPITA DEVELO? PM ARIABLIES AND LOGARITIiU TIE VARIABL3 0F Untransfor3d variablos Yohicles -Telephonos GNP -Doctors ONP ~-Doctr ~ 27 438 ,37 037 .40 168 .69 .30 ,32 '32 q42 38 .69 -Tele phones S33 .22 Energy GNP Doctors ,72 GNP Telephones Vehicles Docto:rs Radios ,,73. .81 072 o70 ,84 087 Newspapers .2 482 ;75 W01 ,81 VehiCle 3 TOlph -, 089 e (W81 o74 .78 '80 54 GNP and Energy ar "per capita" **All variables "per capita." 6 All others are "persons per." )77