f Correlation Measures 0 After studying this chapter, the learners will be able to: + Understand the meaning of the term correlation. + Analyse the degree and direction of relationship be + Calculate correlation through different methods. + Draw Scatter diagram to represent relationship betwe tween two variables. en two variables. ipter Outline a + Introduction + Correlation and Causation ‘= Kinds of Correlation « Degree of Correlation + Methods of Studying Correlation (A) Scatter Diagram. (B) Karl Pearson's Coefficient of Correlation | (© Spearman’s Rank Correlation = List of Formulae INTRODUCTION In the previous chapters we have discussed measures of central tendency (Mean, Median and Mode), partitional values (Quartiles) and measures of dispersion (Range, Quartile Deviation Mean Deviation, Standard Deviation and Lorenz Curve). These are all related to the description and analysis of only single variable. This type of statistical analysis is called ‘univariate analysis ‘Now, we will deal with problems involving association in two variables. We find that in sos ‘as natural sciences, where more than one inter-dependent variables are involved, change as well ses with height, in one variable brings change in others. For instance, weight of a person increas metry we know the circumference of a circle depends on the radius, prices vary wit in geo agricultural production supply, cost of industrial production varies with the cost of raw materials; depends on the rainfall, etc. The relationship between variables is measured by correlation analysis. ‘Thus, ‘the term correlation (or covariation) indicates the velationship betwe which change in the values of one variable, also changes the values of the other variable’ This $# analysis of such data is called bivariate analysis. ren two such variables istical Other Definitions si ‘According to Croxton and Cowden, “When the relationship is of a quantitative nature, the appror” statistical too! for developing and measuring the relationship and expressing it in a brief, formu 5 J as correlation.”ial EE he accompanied by corre cares Measu oh *SPOning movemerng es a es of Correlation hk. ins 3m others) thes petty $9 that movements in on are Said to be correlated.” e “ge geen between {Wo Variables dy aust and effect: ; ‘een two variables, For example, temperature. Si n 2 ase in ture. Similarly, nerease I arinble is affected sinwltgnon 7% 2238 sence Tadius affects the anaes may be a price rise due to change ey large number pe part are rarely found. factors: It is almost impossible to find out eet eon Se For eunple there jn social sciences correlation rarely melee factor is the 'y and so many other 4. Effect of third variable: Sometimes, both thea ihird variable or by more than ong very elation between yicld of rice and vicki cit 2 commoaities have ted atencn cauise of rise in prices. Thus, Se and effet relationship. eS related var mae varibles ae being affected by @ For example, we may find high degree of ‘heat. In reality it may be found that yield affected either by rainfall and other factors like fertilizers, quality of seeds, irrigation ia ol sce ans aa 8 n facilities, etc. There may not be any real relationship heat. I s Be insions In such cases, correlation can give misleading use affecting two variables; W ney tecing we vini e w dhtie n ee Sm Re ic jad riccca caine eee peitol prices are increased, the airfares and taxifare are also increased sie: IL Chance: A high degree of correlation may be derived due to sheer coineidence. In such tases, these variables are not in any way related to exch other and therefore, do not have any relationship. For example, the population of 2 country is increasing and production of sugar is also increasing—it is purely a matter of chance Similarly, number of divorces ; per year and the export of television sets, production ‘of shoes with the agriculture production, félationship between car produced and children born 8 country. This kind of correlation Healled ‘chance or spurious’ coreelation. It should tbe understood clearly that correlation is [ a relationship only between related variables. . The above points. make it clear that correlati’ only.2 ws amined soles and it cs necesearily signify a cause and effect ‘elationship between the varia es alt Bis alco be remembered that existence > correlation isn0 pie a eta ne Beer two variables would always be of the same type. There may tease in price the demand falls, i whenever mere i= an incr ase in pri ee onan Sfeeever ‘here 13.5 Fee Mt pea the demand must Bp cee Bat ts a we the demand may also 60 UP. as seen in case of BEI posible tat wi Fined ot myerene ie ‘and social sciences various lentation goods. This is 0" ‘account o act gant pes 2 . i es co-variation, not causation. ees aE ‘Thus, corelation measures a i of these factors separately , ort eg an Mcssures the direction and intensity ° | : ORRELATION eet es casteat na be: ie elation Xe Fis and Negetve UTp ‘is in opposite directions, that is said to be negative or inverse. For For example: ‘One variable is increasing, the other is decreasing 200 15 60 2 i “We find that in I (a) the values of both X series and y series are increasing, In I (#) values of X and Y are decreasing. Thus, they are both examples of positive correlation. On the other hand, in If (@) the values of X are increasing and the values of Y are decreasing. Similarly, in II (b) the ‘values of X are decreasing and the values of Y are increasing. Thus, they are both examples of negative correlation. Examples: (Positive Correlation) 1. Age of husband and age of wife. 2, Price of commodity and quantity supplied. 3, Increase in cost of advertisement and sales. 4, The family income and expenditure on luxury items. 5, Age of life insurance applicants and premium of life insurance, Examples: (Negative Correlation) 1, Demand of a commodity may go down as a result of rise in prices, 2, Sale of woollen garments and day temperature. 3, Yield of crops and price ‘4, A fall in temperature results in more sale of heaters. 5. A fall in temperature with a rise in the altitude. 8 2. Linear and Curvilinear Correlation If the ratio of change between two variables is uniform then there exists linear correlation between them. Linear and nor-linear correlation is based upon the consistency of the ratio of change between the variables. Their relationship is described by the straight line. If the variables under are graphed, the points will form a straight line. In case of non-linear relationship (curvilinear), the amount of change in one variable does not oo Seonetant ratio 1 the amount of change in the other variable. Such relationship will form pve on grefipie-and po ee of number of variables Correlation: When onty nett is called simple correlation, te tion and rainfall, inom FOr d ae wil be sip er ° tion, ther multiple or partial, py tT MOte variables juction, rainfall and the ie the relati, variables from bath, pana iP, of S49 variables is studied by Renin orrelation is a stud tionship g kept constant. For example, the ©, each other, effect of other induce eae *, the con Sat i relation ang ona 9 Stan ct ep i en od DEGREE OF CORRELAT) ship between two values can by ey tion which is obtained by calcula Correlation: Perfect correlation is that proportional. If equal proportional changes are in the same direction, there is perfect positive correlation yeen the two values described as +1 On the other hand, if equal proportional changes are in the reverse direction, there is perfect gative correlation described as ~1 determined by the quantitative value of coefficient tions, where changes in two related variables are or example, the circumference of a circle increases in the equal proportionate ratio. with the gin the length of its diameter, the amount of electricity bill increase in a perfectly definite o with an increase in the number of units consumed, etc. fe1o Correlation: The value of the coefficient of correlation may be zero, This does not mean Jation between the two variables. Two values are uncorrelated. There other type of relation may exist eabsence of any type of re 9 linear relationship between them however, Degree of Correlation a (an Jegree of Correlation No Correlation Lente — Positive Neon . Low 4 Moderate ows 25 | High + 2510 76 jables may be correlated, Put am social nese te ane go el ine ah of Correlation. ac é ioe when there are not sae limited positive w) . one variable need 90" jon is said to Be ion is limited negative when Ehere ease) in the other variable: COT girection: and s in the two variables 1"a ensures inthe mre dio, Te ie! ST Rate mah +.75 to 1); moderate (+ .25 to.75) or low (between O10 A tionship, tion indicates strong linear relationship and low value indicates we . Thus, the calculated result of correlation must vary between ~ The following chart will show approximate degree of correlation 2 to +1. \ecording to Karl Pearson's formula, [teen ae co Perfect Correlation. +1 y sostt| to siti Sie ea Very high degree of Correlation + Sor more GAM Suificiently high degree of ahs from - 75 to~ $ aes from + 750+ 9 z i Moderate degree of Correlation from+25t0+75 | __—from 025 to 75 i Low degree of Correlation Bae hak 2005 EE Zero Correlation (Uincorrelated) 9 Joo sollte Sain METHODS OF STUDYING CORRELATION There are various methods of finding correlation. Some of them are: (@) Scatter Diagram (b) Karl Pearson's Coefficient of Correlation (c) Spearman's Rank Correlation A. SCATTER DIAGRAM Scatter diagram is a simple visual method for getting some idea about the presence of correlation. It is obtained by plotting values on a graph paper. The chart is prepared by measuring X-variable ‘on horizontal axis and the ¥-variable on vertical axis and plotting a point for cach pair of observation of X and Y values. In this way, the whole data are plotted in the shape of points The cluster of points, so obtained on graph paper is called the scatter diagram. When the plotted points show some trend—upward or downward—we know that there is some correlation between the variables. When the trend is upward (ce, the plotted points lie from lower left comer to upper right corner) the correlation is positive, when it is downward (i.e., the plotted points lie from the upper left corner to lower right corner) the correlation is negative. Let us study the scatter diagrams given below: Figures (a), (b) and (c) show an upward trend, ie., they show positive correlation ol______» "or Dea of Fone Corton Low Degree of Positive Coretation ©(@, (© and (f) show a downward trend, i ee _ “4 | le. they show negative correlation, x gh aa ddida ys, 1 ek High Dgree of Negative Corrolation pie Negative Correlation @ oe Low Degiee of Negative Correlation 0 Sr + We find from the plottings on the scatter diagrams that there is a certain similarity between —J ipand (0, (8) and (e) and (0) and () ‘In figures (2) and (d) the plotted points are almost in straight lines—this indicates perfect comelaton. + In figures (6) and (¢) the plotted points are not in a straight -— ihe but if we draw a straight line in the middle of points { ¥ (cegression line) we will find, the points are nearer to the | Natal line, This kind of scatter diagram shows high degree of correlation. Vall iin figures (€) and () if we draw a similar line (regression line), we will find that the plotted points are very much leone scattered around the line—not as near as in the case of (6) | ee and (¢). This kind of scattered diagram shows low degree of correlation. Finally, figure (g) shows such a vast scatter of points that it is impossible to see any trend—this shows no correlation or zero correlation Fig. 1 Illustration 1. From the following pairs of - ss lation. value of variables X and Y draw a scatter na ial dagram and interpret the result jes, on Kal a Mea 5 6 7 6 9 10 11 12 13 14 18 | a TY: 78 72 66 60 54 48 42 36 30 24 18 122 wes Solution re We note that X = 4 and Y = 78 are givenvas the nts fist X and Y values. We may plot this as point ay ¥) on graph paper, where X = 4 and Y = 78. le measure 4 on X-axis and 78 on Y-axis. Similarly for second values of X and ¥, where X= 5 and Y =72, the coordinates of second point are (5, 72). Wemeasuire 5 along the X-axis and 72 along Y-axis nd so on for all the given X and Y values. From the above scatter diagram we can di that the variables X and Y are correlated: The poin's Tike the shape of a line, and it slopes down from toP ‘perfect negative correlation between X and Y. lecide 7 that there left to bottom right. This showsI statistics for Economics-X! Merits and Demerits of Scatter Diagram Merits : 1, It is very easy to draw a scatter diagram. " 2 Scatter diagram is an attractive and easy method 10 find the nature of correlation between two variables, It is a mon-mathematical and easy method. Tk is not affected by extreme values. a 4. Itis the first step in determining the form of relationship. One ean now whether variables Bre correlated or rot at a glance by the scaiter diagram. It also helps to know whether Variables are positively correlated or negatively correlated, linear or non-linear in case of linear relationship between X and Y, it gives a clear visual picture of the proportionate change in the value of ¥ for a change in the value of X. Demerits 1. The degree of correlation between two variables cannot seatter diagram, 2. It gives a rough idea of how the two variables are related either low or high, but definite conclusions cannot be drawn by just examining the diagram. 3. Scatter diagram does not indicate the direction of causation causes X or X causes Y. 4, If there are more than two variables, then it is not possible to draw a scatter diagram, be known in numerical terms by It does not tell, whether ¥ B. KARL PEARSON'S COEFFICIENT OF CORRELATION A mathematical method for measuring the linear relationship between the variable X and Y was suggested by a great British Bio-metrician and statistician Karl Pearson (1867-1936) It is the most widely used method in practice and is popularly known as Pearsonian correlation of coefficient. This is also called the product moment correlation. It is represented by r. Itis based on arithmetic mean and standard deviation. The Coeffficient of Correlation (x) of two variables is obtained by dividing the sum of the products of the coresponding deviations of the various items of two series from their respective means by the product af their standard deviations and the number of pairs of observations Symbolically, oxy Nxo, xo, Here, x=(X-X); y=(¥-¥) Standard deviation of X Series, ie., fe g, = Standard deviation of Y Series, i. N= Number of pairs of observations "VN r= Coefficient of correlation ‘The above formula can be rewritten as under ixy O,.Dy lyine i prod! a) Actus (p) Direc fo Asst i Step. (a) Actual Tilustrat the result. Serial As. Mark Solutic7 Se or avy : Ext By? Bix 3y fay ig the Karl Pearson’s formula, Coefficient of Correlation is calculated by Storing Mean Method Method sumed Mean Method sp Deviation Method Mean Method tration 2. Calculate Product moment correlation from the following data and interpret It. No. of Shudemts in Mathematics Is Statistics =| 25 | 2 | oa aie ba lution. Karl Pearson's coefficient of correlation is also called Product Moment Correlation. Calculation of Coefficient of Correlation ‘Marks it y-¥ ] \ 2 | Statistics, fa 4 7 | oA 25 2 ee 4 Ss 1 36 oT zr 26 a 36 4 9 z 0 0 35 4 36 - of = 4 a= == Y= SY . 290 33 i 10 fax Zxy = 708, Ux? =1080 and Ly? = 480 708 rs ae 080% 480 Vis400 20 re is high degree of positive correlation between X and Y. Calculate Karl Pearson's coefficient of correlation between birth rate and death following data: ‘Yearee Exy = 81, Ex? = 90 and By? = 86 po Catia mesial Vo0x86 S770 879772 = 0920 e, there is high d sitive in th rate. Hence, there is high degree of positive correlation between birth ‘een birth rate and death ri Mllustration 4. From the following data compute product moment correlation betwee lation between X and Yr X series No. of items ay Arithmetic Mean eB z ~ 18 Square of deviation from Mes " 136 138 Summation of product of deviat x of deviations of X and Y serie i ee ns of X and Y series from their respective Solution. Regarding deviations of the val a values in X and Y series from their respective ‘we are given the following information: i Ex? = 136, y? = 138, Exy = 122 Applying formula: r death Now, we get a Hd = wi Borie | 18768 136.9% cre is high degree of positive correlation between X and ¥- wo series X and Y with 50 items each have sterdard deviations A and 35 ean of product of deviations of X and series Hom, iat Golsagie Find the coefficient of correlation between X and Y- 35 and Ixy = 420 Hence, th Mlustration 5. respectively. If the sum: arithmetic means be 420, Solution, Given N = 50, 6, = 45,0, Applying formula, Now, we get 420 420 9.533 P= pana 785 ive cor X and Y val 's coe gelation between X and ¥. vdables ig +12.9 and variances of X and gree of positi fficient of correlation between them. Hence, there is moderate de; between Illustration 6. If the covariance Yare respectively 13.8 and 164 F Solution. We are given X and ¥ Jind Karl Pearson's Covariance ofon &, Calculate the number of items for which r = + 0.8, Exy = 200, standard d 2x? = 100, where x and y denotes deviation of items from actual mean. Applying formula, Say 200 cone Sea: & ae lsy2 GF N NEXY—2Xay Wine eee proach 2: If arithmetic ic me give accurate or” of X and Y are not in fraction then the follo 3 : m A= — X = 120 i y Bemis = 3- = 30 Correct EX = 120-8 -12+8+ 10-118 Comect EY = 90-10-7+12+8=93 Correct EX? = 600 - 8? — 127 + 8 + 10% = 556 Correct SY? = 250 - 10? - 72+ 12? + 82 Correct EXY = 356 ~ (8 x 10) ~ (12 x 7) + (8 x 12) + (10 x 8) = 368 = 309 NEXY-3xzY 30 368 —118%% yrEe-e 0x56 (118) 30% 08 — 3) _ 11,040 ~ 2 Lies io ¥'16, 680 -13,924 9, V2 756V6R 52.50 «24.92 = © = 0051 13083 a Coefficient of correlation is 0.051 Assumptions of Karl Pearson's Coefficient of Correlation Pearsonian coefficient is based on following assumptions: 1. Linear relationship: If two variables are plotted on a scatter diagram, itis assumed tat the plotted points will form a straight line. Thus, there is a linear relationship betes the variables, ttNormality: The correlate related variables Meauras of Conelcaor ial aa Mpich form a normal di ate aff istriby fected ‘by wands and Wives, height of ea aes by ale numberof independent causes fores and @ normal distribution fathers end tons pee eee a cosa relationship: Correlation ve formed, and demand are affected by such s only meani gf repye is no such Telationship, Th a os isa cause and effect relationship factors that affect these sri. Tete Is-n0 relationship bet eae bi ae ae ween rice and wheat, because the Haring the last ten years ma ae te es are not common Sina he weight of an indivi aay also show similar tendency but en upward trend and hs paler a ecause the forces affecting the ty there cannot be any eorslation be ee eee © time CO ehip berween the arial, whether such zelaonshiP i 5 alineat rejationshiP | 3. there i Itassumes tha seas a very careful interpretation, otherwise exists or not ees Be ereiation lies between” it may be misinterprete350 Ill statistics for Economics-X! C. SPEARMAN’S RANK CORRELATION Sometimes we come across statistical series capable of quantitative measurement but can which the variables under consideration ate n ‘be arranged in serial order. This happens when s) such as honesty, beauty, character iS are dealing with qualitative characteristics (attributes) such as honesty *, Morality, intelligence of students, leadership quality, cooperation, ete. They cannot be measured quantitative (eumerically) but can be arranged serially. In this situation Karl Pearson’s coefficient of correlation cannot be used. Charles Edward Spearman, a British psychologist developed a formala in 194 ‘which results in obtaining the correlation coefficient between ranks of N individuals in the two attributes under study called coefficient of correlation, by rank differences. It is the Product Moment Correlation between the ranks This method is applicable only to individual observations rather than frequency distribution, The result we get from this method is only an approximate one, because under ranking method, original values are not taken into account. ‘After assigning ranks to the various items, the differences of corresponding rank values are calculated and following formula is used: op’ where, ,, = Coefficient of rank correlation ED? = the total of squares of the differences of corresponding ranks N =the number of pairs of observations Like Karl Pearsons coefficient of correlation, the value or r, lies between +1 and -1. If 7, = 41, then there is complete agreement in the order of ranks and the direction of the rank is also the same, When r, = -1, then there is complete disagreement in order of ranks and they are jn opposite direction. Let us examine by following example: Fae | Perfect positive correlation, ie there is 2 | Perfect negative correlation, i.e, there is complete agreement complete disagreement. ‘There are three cases for calculation of rank correlation: (a) When Ranks are given. (b) When Ranks are not given. (c) When Ranks are equal or repeated Nov whahi ank is -y are F when ranks are given; oe fe pedge 1 2 pi 2 2 8 calculate the rank correlation coos " icient, solution. Contestant jm) ola]e=|>| lelele ie if Steps: eo. Wijuatation 16.19 & baby competition {wo judges rank ie bia 3 e 1S ee r 7% Measures of correlation hi. 351 the 11 entries as follows: ° 2 1. Calculate the difference of two ranks, ie, R, ~ R,- Denote these differences by D. 2. Square these differences and find out the total ED, 4, Apply the following fo Now we get, where, mula: =1- 0.091 0.909 ive correlation, of positi Hence, there is high degree Of POSTS. strong likes f 0.909. It indicates that ju ot cecned. ges have jie, two judges are agreeing to the degree and dislikes so far as ranking of theMlustration 17. From the following marks obtained by 10 students in Statist ics and calculate Spearman's coefficient of rank correlation. Beonomics [eens ts l>a le lela = Ezonomics | 50 35 7 | 35 sa] 8 ~ Solution, Calculation of Coefficient of Rank Correlation Statistics % Economies (ae Ey a | 2 : 56 20 65 Illus and ther Steps: 1. Assigns ranks to the given data. Ranks can be given by allotting the biggest item the fint tank, the next to it should be allotted second rank and so on or by allotting the smallest item the first rank, next to it should be allotted second rank and so on. Any one of the Rank above method of ranking must be followed in case of both the variables 2. Find the difference between the two ranks (i.e, R,— R,) and denote these differences by D. 3. Square these differences and find out the total D2 4. Applying the formula, we get ED? ele NPD Here, 2D? = 318 and N = 10 1,927 = -0.927. Thus, there is high degree of negative correlation. Our conclusion would be that rankings obtained by students in Statistics and Economics are highly reverse to each other. It should be noted that the rank correlation (z,), is equal to the product moment consi (Karl Pearson’s method) between the ranks. Let us examine by the ranks obtained in Illustra rerst est he = 0.927, which is same as before Mlustration 18. Ten entries are submitted for a competition. Three judges study each entry and then list the ten in rank order. Their ranking. are as follows: Entry No. 1ja® o 3 brd p SqtnSle Zag Beeman Judge A 9 3 7 5° 1 6 jam eiangenal Ranks given by’ i Fuddge B vietoatl euginival”: 10) bas Asi Seq 2b W75mhe Judge C 44) 6.208 pecs gehien anal Calculate the appropriate rank correlations to help you to answer the following questions: (@) Which pair of judges agree the most? (b) Which pair of judges disagree the most aleulation af Coefficient of Rank Correlation Solution. Ss a Te | oF | OF T Rank D, ¥ 2 pay Ram: by by (ase) | | Co) | oe Roh 4 mete 6 = Judge B fudge © | : G Sieae 154 IML Statistics tor Economies Xt Applving the follewing formula, we get oie he ; NON rs (between judges 4 and 8) eee 28 1 029 2 ie +071 1, fhetween judges A and C3 med am 10 she * 1 1 = DASRS 108125 r, (between Iuciges B and CY aim 2 i" ww 1-053 “a A) Siaos ¥, of the pair of ju un they have the nearest they agree the must (@) Since r, of the pair of j ges B and C, is minimum therefore, th jent of rank following data 19, Caleulate the coef Peedi 1¢ «(16 th CBB ¥ ee 6 15 4 2 ae s D w 25 - 15 y | 15 1 25 y +10 1.0, = +10 10. | 25 05 023 ees a usar I Is easy to caleulate and simple te understand a8 compared to Ke:l Pearstm’s M only the ranks of difforest item-values in the variables are given, Rank Bree, ee for finding the degree of correlation. ala are of qualitative nature like beau! , honesty, intelligence etc., this method De employed usefully. eons i fhe Values ore:not repeated, answer obtained by Karl Pearson's method and Rank p Method will be the sam a is An case of actual data, rank method can be sed lo get rough degree of conmlation. Demerits: 1, This method is applicable only to andividual observations, rathet than frequency d his method cannot be employed for finding out correlation in a grouped distribution (bivariate disteibution). cane - If the number of jleats are more than 30, if becomes difficult to find out vanks amd thett differences. Therefore, it is not advisable to use this method wher. number is 30, unless ranks ee given to us. In such eases, Karl Peatson’s method is more suitable 3. Tt is based only on ranks, as all the information relating to data is not used, this lacks precision as compared to Karl Pearson's method. LIST OF FORMULAE tion son's Co | Mean Method 1, Karl Pe: (a) Actual