SD vs. SD + One of the most important uses of sample statistics is to estimate the corresponding population parameters.

Similar documents
The standard deviation 1 n

Normal Distributions

( n+1 2 ) , position=(7+1)/2 =4,(median is observation #4) Median=10lb

UNIT 4 Section 8 Estimating Population Parameters using Confidence Intervals

Intermediate Statistics

OCR Statistics 1. Working with data. Section 3: Measures of spread

Polynomial Functions and Models. Learning Objectives. Polynomials. P (x) = a n x n + a n 1 x n a 1 x + a 0, a n 0

Arithmetic Sequences

Fundamentals of Media Processing. Shin'ichi Satoh Kazuya Kodama Hiroshi Mo Duy-Dinh Le

SAMPLE VERSUS POPULATION. Population - consists of all possible measurements that can be made on a particular item or procedure.

The isoperimetric problem on the hypercube

ENGI 4421 Probability and Statistics Faculty of Engineering and Applied Science Problem Set 1 Descriptive Statistics

Performance Plus Software Parameter Definitions

Bezier curves. Figure 2 shows cubic Bezier curves for various control points. In a Bezier curve, only

1.2 Binomial Coefficients and Subsets

EVALUATION OF TRIGONOMETRIC FUNCTIONS

IMP: Superposer Integrated Morphometrics Package Superposition Tool

EM375 STATISTICS AND MEASUREMENT UNCERTAINTY LEAST SQUARES LINEAR REGRESSION ANALYSIS

condition w i B i S maximum u i

Improving Template Based Spike Detection

Data Analysis. Concepts and Techniques. Chapter 2. Chapter 2: Getting to Know Your Data. Data Objects and Attribute Types

Area As A Limit & Sigma Notation

Alpha Individual Solutions MAΘ National Convention 2013

Parabolic Path to a Best Best-Fit Line:

The Closest Line to a Data Set in the Plane. David Gurney Southeastern Louisiana University Hammond, Louisiana

Math 167 Review for Test 4 Chapters 7, 8 & 9

Descriptive Statistics Summary Lists

Image Segmentation EEE 508

PLEASURE TEST SERIES (XI) - 04 By O.P. Gupta (For stuffs on Math, click at theopgupta.com)

Lecture 28: Data Link Layer

Vision & Perception. Simple model: simple reflectance/illumination model. image: x(n 1,n 2 )=i(n 1,n 2 )r(n 1,n 2 ) 0 < r(n 1,n 2 ) < 1

CHAPTER IV: GRAPH THEORY. Section 1: Introduction to Graphs

Designing a learning system

Consider the following population data for the state of California. Year Population

Lecture 5. Counting Sort / Radix Sort

Aberrations in Lens & Mirrors (Hecht 6.3)

Name Date Hr. ALGEBRA 1-2 SPRING FINAL MULTIPLE CHOICE REVIEW #1

DATA MINING II - 1DL460

Math 10C Long Range Plans

Learning to Shoot a Goal Lecture 8: Learning Models and Skills

Chapter 5. Functions for All Subtasks. Copyright 2015 Pearson Education, Ltd.. All rights reserved.

The golden search method: Question 1

Pattern Recognition Systems Lab 1 Least Mean Squares

Administrative UNSUPERVISED LEARNING. Unsupervised learning. Supervised learning 11/25/13. Final project. No office hours today

9.1. Sequences and Series. Sequences. What you should learn. Why you should learn it. Definition of Sequence

Lecture 1: Introduction and Strassen s Algorithm

New Results on Energy of Graphs of Small Order

COMP 558 lecture 6 Sept. 27, 2010

Assignment 5; Due Friday, February 10

What is the difference between a statistician and a mortician? Nobody's dying to see the statistician! Chapter 8 Interval Estimation

Numerical Methods Lecture 6 - Curve Fitting Techniques

12-5A. Equivalent Fractions and Decimals. 1 Daily Common Core Review. Common Core. Lesson. Lesson Overview. Math Background

. Written in factored form it is easy to see that the roots are 2, 2, i,

CS Polygon Scan Conversion. Slide 1

Python Programming: An Introduction to Computer Science

Name Date Hr. ALGEBRA 1-2 SPRING FINAL MULTIPLE CHOICE REVIEW #2

Designing a learning system

Appendix D. Controller Implementation

Wavelet Transform. CSE 490 G Introduction to Data Compression Winter Wavelet Transformed Barbara (Enhanced) Wavelet Transformed Barbara (Actual)

Module 8-7: Pascal s Triangle and the Binomial Theorem

1 Graph Sparsfication

Recursive Procedures. How can you model the relationship between consecutive terms of a sequence?

Ones Assignment Method for Solving Traveling Salesman Problem

Computers and Scientific Thinking

1. Sketch a concave polygon and explain why it is both concave and a polygon. A polygon is a simple closed curve that is the union of line segments.

Mathematics and Art Activity - Basic Plane Tessellation with GeoGebra

Analysis of Server Resource Consumption of Meteorological Satellite Application System Based on Contour Curve

Xbar/R Chart for x1-x3

Ch 9.3 Geometric Sequences and Series Lessons

Investigation Monitoring Inventory

CSC 220: Computer Organization Unit 11 Basic Computer Organization and Design

Capability Analysis (Variable Data)

CS 111: Program Design I Lecture #26: Heat maps, Nothing, Predictive Policing

CS : Programming for Non-Majors, Summer 2007 Programming Project #3: Two Little Calculations Due by 12:00pm (noon) Wednesday June

Morgan Kaufmann Publishers 26 February, COMPUTER ORGANIZATION AND DESIGN The Hardware/Software Interface. Chapter 5

South Slave Divisional Education Council. Math 10C

ECE4050 Data Structures and Algorithms. Lecture 6: Searching

Intro to Scientific Computing: Solutions

Accuracy Improvement in Camera Calibration

The number n of subintervals times the length h of subintervals gives length of interval (b-a).

K-NET bus. When several turrets are connected to the K-Bus, the structure of the system is as showns

Apparent Depth. B' l'

Lecture 2: Spectra of Graphs

Algorithms for Disk Covering Problems with the Most Points

Lower Bounds for Sorting

Our Learning Problem, Again

Analysis Metrics. Intro to Algorithm Analysis. Slides. 12. Alg Analysis. 12. Alg Analysis

A Taste of Maya. Character Setup

Optimal Mapped Mesh on the Circle

ENGR Spring Exam 1

Counting II 3, 7 3, 2 3, 9 7, 2 7, 9 2, 9

Creating Exact Bezier Representations of CST Shapes. David D. Marshall. California Polytechnic State University, San Luis Obispo, CA , USA

A MODIFIED APPROACH FOR ESTIMATING PROCESS CAPABILITY INDICES USING IMPROVED ESTIMATORS

Computational Geometry

1. The lines intersect. There is one solution, the point where they intersect. The system is called a consistent system.

Basic Design Principles

BAAN IVc/BaanERP. Conversion Guide Oracle7 to Oracle8

PRESENTER DISCLOSURE MEASURING HEALTH INFORMATION TECHNOLOGY USE AND EHEALTH LITERACY AMONG AFRICAN AMERICANS BACKGROUND

Counting Regions in the Plane and More 1

islerp: An Incremental Approach to Slerp

Transcription:

SD vs. SD + Oe of the most importat uses of sample statistics is to estimate the correspodig populatio parameters. The mea of a represetative sample is a good estimate of the mea of the populatio that the sample represets. The SD of a represetative sample teds to uderestimate the SD of the populatio from which it was draw. To correct for this, statisticias use the SD + of the sample to estimate the SD of the populatio. If is sample size, the 1 SD + = 1 SD sample = (xj x) 1 2 If the sample size is large, the there is o sigificat differece betwee SD ad SD + because /( 1) 1 whe is large. The SD + is called the sample stadard deviatio. 1

Questio: How is the data clustered aroud the mea? The aswer is easiest to give i terms of the stadard deviatio: I ay data set, the proportio of the data that lies more tha k SDs from the mea is always less tha 1/k 2. This fact is kow as Chebyshev s iequality, ad follows directly from how the stadard deviatio is defied (with a little work ad cleveress). For example, less tha 1/4 = % of the values i ay data set lie more tha 2 SDs from the average value (mea). Less tha 1/9 11.11% of the data lie more tha 3 SDs from the average value. Etc. Or we ca say that i ay data set, more tha 75% of the data lie withi 2 SDs of the mea, ad more tha 88.88% of the data lie withi 3 SDs of the mea. The estimates above are true for ay set of data. O the other had, if we kow more about the data, the we ca ofte get sharper estimates. 2

For certai types of data sets, almost all of the data lies withi two or three SDs of the average. Example (from the book): h = 63.5 iches ad SD h 3 iches... 3 Statistics, Fourth Editio Copyright 2007 W. W. Norto & Co., Ic.

Example (cotiued): h = 63.5 iches ad SD h 3 iches... Statistics, Fourth Editio Copyright 2007 W. W. Norto & Co., Ic. 4

To better uderstad clusterig aroud the mea, we itroduce Stadard uits If x j comes from data with average x ad stadard deviatio SD x, we covert x j to its stadard uit, z j, by settig z j = x j x SD x. (*) z j tells us how far x j is from x as a multiple of SD x. (*) If z j > 0, the x j is above average; if z j < 0, the x j is below average. (*) Stadard uits are pure umbers : there are o uits of measuremet (iches, dollars, etc.) associated with stadard uits. The uits of the x j s are cacelled whe we divide x j x by SD x. (*) The stadard uit value z j of a give datum x j is also called the z-score of x j. 5

Example. Suppose that the average Jauary temperature i Poduk is 45 F, with a SD of 2 F, while i Whoville the average Jauary temperature is F with a SD of 5 F. O Jauary 20th, the temperature i Whoville was 16 F ad i Poduk it was 38 F. Where was the temperature more uusual that day? We ca aswer this by covertig the temperatures o Jauary 20th i both tows to stadard uits: z p = 38 45 2 = 3.5 ad z w = (*) Both temperatures were below average. 16 5 = 1.8. (*) The z-score for Poduk is more egative tha the z-score for Whoville, so from a statistical poit of view the temperature i Poduk was more uusual that day. Ituitio: The larger z j, the more uusual x j is (i the set of data it came from). 6

Observatio. Covertig ay set of data, {x 1, x 2,..., x } with average x ad stadard deviatio SD x = s, to stadard uits produces a set of umbers {z 1, z 2,..., z } with average z = 0 ad stadard deviatio SD z = 1. Because arithmetic... z = z 1 + z 2 + + z = = = x 1 x s x 1 x + x 2 x s + x 2 x x 1 +x 2 + +x = x x s = 0 s + + x x s + + x x s {}}{ x + x + x 7

ad more arithmetic SD z = = = = = z 2 1 + z 2 2 + + z2 ( x1 x s ) 2 ( + x2 x s ) 2 ( + + x ) x 2 s (x 1 x) 2 s + (x 2 x) 2 2 s + + (x x) 2 2 s 2 (x 1 x) 2 +(x 2 x) 2 + +(x x) 2 s 2 (x 1 x) 2 +(x 2 x) 2 + +(x x) 2 s 2 = s s = 1 8

The ormal approximatio, I Differet sets of data may be see to have very similar distributios, oce they have bee coverted to stadard uits. Covertig to stadard uits moves the ceter of the histogram (the average of the data) to 0, ad scales the data as a whole so that oe SD is coverted to 1 uit. I may cases, the histogram of the data, oce coverted to stadard uits, takes o a somewhat bell-shaped form the form of the ormal curve. The ormal curve is the graph of the fuctio (where e = 2.7182818...). y = 1 2π e z2 /2, 9

50 5-4 -3-2 -1 0 1 2 3 4 z The ormal curve is symmetric aroud the lie z = 0, ad the total area uder the curve is equal to 1 (or 100%, if you prefer). 10

Example: The distributio of heights of wome age 18 ad over i HANES5 (Health ad Nutritio Examiatio Study, 03-04) appears i the histogram below (from page 81 i chapter 5 of FPP). The average height is 63.5 ad the SD is about 3. The shaded regio represets the heights that fall withi oe SD of average. 11

To see how well the distributio of the height data is approximated by the ormal curve, we must covert the data to stadard uits ad sketch the histogram for the stadardized (or ormalized) data. To save a lot of drawig time... otice that the coversio to stadard uits is just a rescalig. This meas that istead of covertig all of the heights to their stadard uits ad the drawig a ew histogram, we ca istead chage the horizotal ad vertical scales o the origial histogram. 12

13

If the (rescaled) histogram is well-approximated by the ormal curve, the area of regios uder the histogram will be approximately equal to areas uder the ormal curve for the same rage of stadard uits. I.e., the percetage of the data that lies withi 1 SD of the average will be approximately equal to the area uder the ormal curve betwee -1 ad 1; the percetage of the data lyig withi 2 SDs of the average will be approximately equal to the area uder the ormal curve betwee -2 ad 2; ad so forth. This is useful, because the distributio of the area uder the ormal curve is well-uderstood. I particular... 14

50 68% 5-4 -3-2 -1 0 1 2 3 4 z (*) The area uder the ormal curve betwee 1 ad 1 is 0.68 = 68%. 15

50 95% 5-4 -3-2 -1 0 1 2 3 4 z (*) The area uder the ormal curve betwee 2 ad 2 is 0.95 = 95%. 16

50 99% 5-4 -3-2 -1 0 1 2 3 4 z (*) The area uder the ormal curve betwee 3 ad 3 is 0.99 = 99%. 17

Rule of thumb : If a set of data has a approximately ormal distributio, the: About 68% of the data lies withi oe SD of average; About 95% of the data lies withi two SDs of average; About 99% of the data lies withi three SDs of average; Remember: This rule oly applies to data that is (approximately) ormally distributed! Abset that coditio (or assumptios about how the data is distributed) we rely o weaker (but more geeral) estimates (like Chebychev s iequality). To calculate areas uder the ormal curve for regios other tha those above ( 1 to 1, 2 to 2 ad 3 to 3), we use a ormal table, like the oe foud i the back of the textbook. 18

A ormal table 19

(From Statistics, 4th ed., W.W.Norto & Co., Ic.) Copyright 200 20

Usig the ormal table (i) The table i the appedix gives the areas for symmetric regios z 0 z z 0 (as percetages), where 0 z 0 4.45. If z 0 4.50, you ca assume that the correspodig area is 99.9999%. Example: Suppose that the heights of me aged 35 i a certai city are distributed (approximately) ormally with a average of 67 iches ad a stadard deviatio of 2.5 iches. What percetage of these me are betwee 65 ad 69 iches tall? a. A height of 65 iches correspods to 65 67 2.5 = 0.8 stadard uits, ad 69 iches correspods to 69 67 2.5 = 0.8 stadard uits. b. The percetage we wat is (approximately) equal to the area uder the ormal curve betwee 0.8 ad 0.8 which is equal to the table etry for z 0 = 0.8, which is 57.63%. 21

(ii) The ormal curve is symmetric aroud z = 0 so the area uder the curve betwee 0 ad z 0 is equal to the area uder the curve betwee z 0 ad 0, ad both are equal to exactly oe half the table etry for z 0. 50 50 = -4-3 -2-1 0 1 2 3 4-4 -3-2 -1 0 1 2 3 4 z 0 -z 0 Example. What percetage of the me i the previous example are betwee 67 ad 70 iches tall? a. 67 iches is average which correspods to 0 stadard uits ad 70 iches correspods to 70 67 2.5 = 1.2 stadard uits. b. The percetage we wat is (approximately) equal to the area uder the ormal curve betwee 0 ad 1.2 which is equal to half the table etry for z 0 = 1.2. This is 76.99/2% 38.5%. 22

(iii) If z 0 > 0, the the area uder the ormal curve to the left of z 0 is equal to 50% plus half the table etry for z 0, because... 50 = -4-3 -2-1 0 1 2 3 4 z 0 50 50 + = 50% + 1 2 Table(z 0). -4-3 -2-1 0 1 2 3 4-4 -3-2 -1 0 1 2 3 4 z 0 z 0 Example. What percetage of the me i the previous examples are less tha six feet, two iches tall? Six feet, two iches is 74 iches which correspods to 74 67 2.5 = 2.8 stadard uits. The table etry for 2.8 is 99.49%, so the percetage of me who are uder 74 iches tall is 50% + 99.49% 2 99.75%. 23

(iv) If z 0 > 0, the the area uder the ormal curve to the right of z 0 is equal to 50% half the table etry for z 0, because 50 = -4-3 -2-1 0 1 2 3 4 z 0 50 50 = 50% 1 2 Table(z 0). -4-3 -2-1 0 1 2 3 4-4 -3-2 -1 0 1 2 3 4 z 0 z 0 Example. What percetage of the me are taller tha 68 iches? 68 iches correspods to 68 67 2.5 = 0.4, so the percetage of me who are more tha 68 iches tall is (approximately) 50% 31.08% 2 = 34.46%. 24

(*) The areas of other types of regios uder the ormal curve ca be calculated from the table by usig (i) (iv) ad the symmetry of the ormal curve aroud 0. For example, if 0 < z 0 < z 1, the the area uder the ormal curve betwee z 0 ad z 1 is because = 1 2 Table(z 1) 1 2 Table(z 0) 50 = -4-3 -2-1 0 1 2 3 4 z 0 z 1 50 50-4 -3-2 -1 0 1 2 3 4-4 -3-2 -1 0 1 2 3 4 z 1 z 0