Mathematics IV — Unit 3

Statistical Techniques I · BAS303/BAS403 · Complete Notes

Central Tendency | Moments | Skewness & Kurtosis | Curve Fitting | Correlation | Regression

1. Central Tendency — Measures of Average

Arithmetic Mean Median (Grouped) Mode (Grouped)

AM = Sfi*xi / Sfi l + (N/2 - c)/f * i l + (f-f-1)/(2f-f-1-f1)*i
Shortcut: a + Sfd/Sf l=lower lim, c=prev CF f-1=before, f1=after

Geometric Mean Empirical Relation

G = (x1*x2*...*xn)^(1/n) Mean - Mode = 3[Mean - Median]
Harmonic Mean Shortcut AM
1/H = (1/n)(1/x1+1/x2+...+1/xn) AM = a + Sfd/Sf (d = x - a)

Grouped Median = l + [(N/2 - c) / f] * i | N = Total freq, c = CF above median

class

2. Dispersion — Range, MD, SD, Variance, CV

Range Mean Deviation Std Deviation

R = L - S (Largest - MD = Sf|xi - x_bar| / Sf sigma = sqrt(Sf(x-x_bar)^2
Smallest) / Sf)
Variance = sigma^2

Shortcut SD: sigma = sqrt[ Sfd^2/N - (Sfd/N)^2 ]

Coefficient of Variation (CV) = (sigma / x_bar) * 100 | sigma^2 = mu_2 = 2nd

moment about mean

3. Moments — Teen Types (3 Types)

About Mean (mu_r) Raw Moment (mu_r') About Origin (V_r)

mu_r = (1/N) * mu'_r = (1/N) * V_r = (1/N) * Sf_i * xi^r
Sf_i(xi-x_bar)^r Sf_i(xi-a)^r V_1 = x_bar
mu_1=0 | mu_2=sigma^2 a = assumed mean

Relations: Raw Moments → Central Moments

mu'_1 = x_bar - a | mu_2 = mu'_2 - (mu'_1)^2

mu_3 = mu'_3 - 3*mu'_2*mu'_1 + 2*(mu'_1)^3

mu_4 = mu'_4 - 4*mu'_3*mu'_1 + 6*mu'_2*(mu'_1)^2 - 3*(mu'_1)^4

About Origin: V_1=x_bar | V_2=mu_2+x_bar^2 | V_3=mu_3+3*mu_2*x_bar+x_bar^3

Trick: mu_1 always = 0 (about mean). Raw mu'_1 = x_bar - a. Exam mein ye yaad karo!

4. Skewness & Kurtosis

Karl Pearson's Sk beta & gamma Coefficients

Sk = (Mean - Mode) / sigma = 3(Mean - beta_1 = mu_3^2 / mu_2^3 [Skewness]
Median) / sigma Range: -3 to +3 Sk=0: beta_2 = mu_4 / mu_2^2 [Kurtosis] gamma_1
Symmetric | Sk>0: +ve | Sk<0: -ve = +/- sqrt(beta_1) gamma_2 = beta_2 - 3
Bowley's Coefficient Kurtosis Types
(Q3 + Q1 - 2*Median) / (Q3 - Q1) Range: beta_2 = 3 --> Mesokurtic (Normal) beta_2
-1 to +1 Use for open-end or unequal CI > 3 --> Leptokurtic beta_2 < 3 -->
Platykurtic gamma_2 = 0 --> Normal

Kurtosis = shape of frequency distribution. gamma_1 = Karl Pearson's coefficient of skewness.

5. Curve Fitting — Method of Least Squares

Curve Type Normal Equations Trick / Transform

y = a + bx (Straight) Sy = na + bSx Sxy = aSx + bSx^2 Direct — no transform

y = a + bx + cx^2 Sy=na+bSx+cSx^2 Sxy=aSx+bSx^2+cSx^3 3 equations, solve for a,b,c

(Parabola) Sx^2y=aSx^2+bSx^3+cSx^4

y = a + bx^2 Put X=x^2 => y=a+bX Sy=na+bSX Substitute X=x^2 first

SXy=aSX+bSX^2

y = ax + b/x Sxy = aSx^2 + nb S(y/x) = na + Derive via dS/da=0, dS/db=0

bS(1/x^2)

y = ax + bx^2 Sxy = aSx^2 + bSx^3 Sx^2y = aSx^3 + Divide by x first

bSx^4

y = ab^x (Exponential) log y = log a + x*log b Y = A + Bx Take log10, A=log a, B=log b

SY=nA+BSx | SxY=ASx+BSx^2 Antilog at end

y = ae^(bx) ln y = ln a + bx Y = A + bx SY=nA+bSx Take natural log (ln) A = ln a =>

| SxY=ASx+bSx^2 a = e^A

6. Correlation — Karl Pearson & Spearman Rank

Covariance: Cov(x,y) = S(xi*yi)/n - (Sxi/n)*(Syi/n)

Karl Pearson r = SXY / sqrt(SX^2 * SY^2) where X=x-x_bar, Y=y-y_bar

 General Formula: r = [N*Sx'y' - Sx'*Sy'] / [sqrt(N*Sx'^2-(Sx')^2) *
sqrt(N*Sy'^2-(Sy')^2)]

Grouped Data: r = [N*SfX'Y'-(SfX')(SfY')] / [sqrt(N*SfX'^2-(SfX')^2) *

sqrt(N*SfY'^2-(SfY')^2)]

Spearman's Rank Correlation

r = 1 - 6*Sd^2 / [n(n^2 - 1)] (d = R1 - R2, difference of ranks)

With ties: r = 1 - 6[Sd^2 + (1/12)(m1^3-m1) + (1/12)(m2^3-m2)+...] / n(n^2-1)

Tie handling: Same value wale items ko average rank do. E.g., 5th & 6th same => rank 5.5. Next rank = 7.

7. Regression — Lines of Best Fit

Regression Line y on x (predict y) Regression Line x on y (predict x)

y - y_bar = byx*(x - x_bar) byx = x - x_bar = bxy*(y - y_bar) bxy =
r*(sigma_y/sigma_x) = SXY / SX^2 r*(sigma_x/sigma_y) = SXY / SY^2
byx Formula bxy Formula
byx = r*(sigma_y/sigma_x) = SXY/SX^2 bxy = r*(sigma_x/sigma_y) = SXY/SY^2

r = sqrt(byx * bxy) | r = SXY / (N*sigma_x*sigma_y)

Property Statement

Same sign bxy, byx and r always have same sign

Product sqrt(bxy * byx) <= 1

Sum bxy + byx >= 2r

Independence Regression coefficients are independent of origin but NOT of scale

Angle tan(theta) = (1-r^2)/r * sigma_x*sigma_y/(sigma_x^2+sigma_y^2)

Intersection Both regression lines pass through (x_bar, y_bar)

Mind Map — Unit 3 Overview

Statistical Techniques I

AM = a+Sfd/Sf | Median = l+(N/2-c)/f*i | Mode = l+(f-f-1)/(2f-f-1-f1)*i | GM, HM |

Central Tendency Mean-Mode=3[Mean-Median]

Range=L-S | MD=Sf|x-x_bar|/Sf | SD=sqrt(Sfd^2/N-(Sfd/N)^2) | Variance=sigma^2 |

Dispersion CV=(sigma/x_bar)*100

About Mean: mu_r = (1/N)Sf(x-x_bar)^r | Raw (mu'_r): a=assumed mean | About Origin:
Moments (3 Types) V_r | Relations: mu via mu'
 Pearson's Sk=(Mean-Mode)/sigma | Bowley=(Q3+Q1-2M)/(Q3-Q1) |
Skewness & Kurtosis beta_1=mu_3^2/mu_2^3 | beta_2=mu_4/mu_2^2 | gamma_2=beta_2-3

y=a+bx (straight) | y=a+bx+cx^2 (parabola) | y=abx => log transform | y=ae^bx => ln
Curve Fitting transform | y=ax+b/x

Covariance formula | Karl Pearson's r | Grouped data formula | Spearman Rank

Correlation r=1-6Sd^2/n(n^2-1) | Tie correction

y on x: byx=r*sigma_y/sigma_x | x on y: bxy=r*sigma_x/sigma_y | r=sqrt(bxy*byx) | Both

Regression lines => (x_bar,y_bar)

Question Solve Karne Ka Shortcut (4-Step Techniques)

1. Central Tendency / Dispersion Questions

1 Table banao: x, f, mid-val, d=(x-a), fd, fd^2, CF columns

2 Summation nikalo: Sf, Sfd, Sfd^2 calculate karo

3 Formula apply karo — AM=a+Sfd/Sf, SD=sqrt(Sfd^2/N-(Sfd/N)^2)

4 CV/Skewness — step 3 results use karo directly

2. Moments Questions (mu' => mu => beta)

1 Raw moments nikalo: mu'_1=Sfd/N, mu'_2=Sfd^2/N, mu'_3=Sfd^3/N, mu'_4=Sfd^4/N

2 Central moments convert karo: mu_2=mu'_2-(mu'_1)^2, mu_3 formula, mu_4 formula

3 beta_1 = mu_3^2 / mu_2^3 (skewness) aur beta_2 = mu_4 / mu_2^2 (kurtosis)

4 gamma_1 = +/- sqrt(beta_1) aur gamma_2 = beta_2 - 3 (excess kurtosis)

3. Regression Lines Given => Find r, sigma

1 Both lines pass through (x_bar, y_bar) => simultaneous equations se x_bar, y_bar nikalo

2 Express as x = f(y) => bxy identify karo | Express as y = f(x) => byx identify karo

3 r = sqrt(bxy * byx) — check: r <= 1, agar nahi toh lines ki assignment swap karo!

4 r * (sigma_y/sigma_x) = byx => sigma_y nikalo (agar sigma_x given ho)

4. Curve Fitting — Kaunsa Formula Use Karen?

1 Equation dekho: Linear? => Direct normal eqns. Non-linear? => Substitute/transform pehle

2 y=ab^x => log10 lo | y=ae^(bx) => ln lo | y=ax^2+b => X=x^2 rakh do

3 Normal equations likh ke table banao => Summation values compute karo

4 a, b (and c) solve karo => original equation mein substitute karo => Final answer!

Last Minute Cheatsheet — Sab Important Formulas

Topic Formula Key Note

AM Shortcut x_bar = a + Sfd/Sf a = assumed mean

SD Shortcut sigma = sqrt[Sfd^2/N - (Sfd/N)^2] Faster than direct method

Pearson Corr. r = [NSx'y'-Sx'Sy']/[sqrt(NSx'^2-(Sx')^2)*sq x'=x-a, y'=y-b

rt(NSy'^2-(Sy')^2)]

Spearman's Rank r = 1 - 6Sd^2 / n(n^2-1) With tie: add (m^3-m)/12

Pearson's Sk Sk = (Mean - Mode) / sigma Or 3(Mean-Median)/sigma

Bowley's Coeff. (Q3+Q1-2M) / (Q3-Q1) For open-end/unequal CI

beta_1 beta_1 = mu_3^2 / mu_2^3 gamma_1 = +/-sqrt(beta_1)

(Skewness)

beta_2 (Kurtosis) beta_2 = mu_4 / mu_2^2 gamma_2 = beta_2 - 3

Regression byx byx = r*sigma_y/sigma_x = SXY/SX^2 y on x line slope

Regression bxy bxy = r*sigma_x/sigma_y = SXY/SY^2 x on y line slope

r from b values r = sqrt(byx * bxy) Check: r <= 1 always!

Normal Eqns Sy=na+bSx | Sxy=aSx+bSx^2 n = no. of data points

(y=a+bx)

y=ab^x fitting Y=A+Bx | A=log a, B=log b Antilog at final step

Exam Tip: Agar r > 1 aaye => lines ki assignment galat hai => SWAP karo! Both lines always pass through
(x_bar, y_bar).

Tie in Rank Correlation: m-th repeat ke liye add (m^3-m)/12 inside the bracket in Sd^2.

Curve Fitting Golden Rule: Agar equation non-linear hai => pehle transform karo (log/ln/substitution), THEN
normal equations likho.

Mathematics IV — Unit 3: Statistical Techniques I | BAS303/BAS303H/BAS403/BAS403H | Bitwise Learning