P. 1
Standard Deviation

# Standard Deviation

4.0

|Views: 484|Likes:

See more
See less

10/25/2012

# Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

Standard deviation

This page may be too technical for a general audience. Please help improve the page by providing more context and better explanations of technical details, even for su bjects that are inherently technical. Standard deviation is the measu rement of the distribu tion of data abou t a mean valu e. It describes the dispersion of data on either side of a mean valu e. A low standard deviation indicat es that the data set is clu stered arou nd the mean valu e, whereas a high standard deviation indicates that the data is widely spread with significantly higher/lower figu res than the mean. T he mean is the arithmetic average.
[1] the standard deviation Formu lated by Francis Galton in the late 1860s, remains the most common measu re of statistical dispersion, measu ring how widely spread the valu es in a data set are. If many data points are close to the mean, then the standard deviation is small; if many data points are far from the mean, then the standard deviation is large. If all data valu es are equ al, then the standard deviation is zero. A u sefu l property of standard deviation is that, u nlike variance, it is expressed in the same u nits as the data.

When only a sample of data from a popu lation is available, the popu lation standard deviation can be estimated by a modified standard deviation of the sample, explained below.

1 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

Contents
1 Definition and calcu lation 1.1 Probability distribu tion or random variable 1.2 Continu ou s random variable 1.3 Discrete random variable or data set 1.3.1 Example 1.3.2 Simplification of the formu la 1.4 Estimating popu lation standard deviation from sample standard deviation 1.5 Properties of st andard deviation 2 Interpretation and application 2.1 Application examples 2.1.1 Weather 2.1.2 Sports 2.1.3 Finance 2.2 Geometric interpretation 2.3 Chebyshev's inequ ality 2.4 Ru les for normally distribu ted data 3 Relationship between standard deviation and mean 4 Rapid calcu lation methods 5 See also 6 References 7 External links

A data set with a mean of 50 (shown in blue) and a standard deviation (σ) of 20.

Definition and calculation
Probability distribution or random variable
T he standard deviation of a (u nivariate) probability distribu tion is the same as that of a random variable having that distribu tion. T he standard deviation σ of a real-valu ed random variable X is defined as:

2 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

where E(X) is the expected valu e of X. E(X) is another name for the mean, and it is often indicated with the Greek letter μ. Not all random variables have a standard deviation, since these expected valu es need not exist. For example, the standard deviation of a random variable which follows a Cau chy distribu tion is u ndefined becau se its E(X) is u ndefined.

Continuous random variable
Continu ou s distribu tions u su ally give a formu la for calcu lating the standard deviation as a fu nct ion of the parameters of the distribu tion. In general, the standard deviation of a continu ou s real-valu ed random variable X with probability density fu nction p(x) is

where

and where the integrals are definite integrals taken for x ranging over the range of X.

Discrete random variable or data set
T he standard deviation of a discrete random variable is the root-mean-squ are (RMS) deviation of its valu es from the mean. If the random variable X takes on N valu es (which are real nu mbers) with equ al probability, then its standard deviat ion σ can be calcu lated as follows: 1. 2. 3. 4. 5. Find the mean, , of the valu es. For each valu e xi calcu late its deviat ion ( ) from the mean. Calcu late the squ ares of these deviat ions. 2 Find the mean of the squ ared deviations. T his qu antity is t he variance σ . T ake the squ are root of the variance.

T his calcu lation is described by the following formu la:

3 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

where

is the arithmetic mean of the valu es x i, defined as:

If not all valu es have equ al probability, bu t the probability of valu e xi equ als pi, the standard deviation can be compu ted by: and

where

and N' is the nu mber of non-zero weight elements. T he standard deviation of a data set is the same as that of a discrete random variable that can assu me precisely the valu es from the data set, where the point mass for each valu e is proportional to its mu ltiplicity in the data set. Example Su ppose we wished to find the standard deviation of the data set consisting of the valu es 3, 7, 7, and 19. Step 1: find the arithmetic mean (average) of 3, 7, 7, and 19,

Step 2: find the deviation of each nu mber from the mean,

4 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

Step 3: squ are each of the deviations, which amplifies large deviations and makes negative valu es positive,

Step 4: find the mean of those squ ared deviat ions,

Step 5: take the non-negative squ are root of the qu otient (converting squ ared u nits back to regu lar u nits),

So, the standard deviation of the set is 6. T his example also shows that, in general, the standard deviation is different from the mean absolu te deviation (which is 5 in this example). Note that if the above data set represented only a sample from a greater popu lation, a modified standard deviat ion wou ld be calcu lated (explained below) to estimate the popu lation standard deviation, which wou ld give 6.93 for this example. Simplification of the formula T he calcu lation of the su m of squ ared deviations can be simplified as follows:

5 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

Applying this to the original formu la for standard deviation gives:

Estimating population standard deviation from sample standard deviation
In the real world, finding the standard deviation of an ent ire popu lation is u nrealistic except in certain cases, su ch as standardized testing, where every member of a popu lation is sampled. In most cases, the st andard deviation is estimated by examining a random sample taken from the popu lation. Using the definition given above for a data set and applying it to a small or moderately-sized sample resu lts in an estimate that tends to be too low. T he most common measu re u sed is an adju sted version, the sample standard deviation, which is defined by

where is the sample and is the mean of the sample. T he denominator N − 1 is the nu mber of degrees of freedom in the vector .
2 T he reason for this definition is that s is an u nbiased estimator for the

6 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

variance σ 2 of the u nderlying popu lation, if that variance exists and the sample valu es are drawn independent ly with replacement. However, s is not an u nbiased estimator for the standard deviation σ; it tends to u nderestimate the popu lation standard deviation. Althou gh an u nbiased estimator for σ is known when the random variable is normally distribu ted, the formu la is complicated and amou nts to a minor correction: see Unbiased estimation of standard deviation. Moreover, u nbiasedness, in this sense of the word, is not always desirable; see bias of an estimator. Another estimator sometimes u sed is the similar expression

T his form has a u niformly smaller mean squ ared error than does the u nbiased estimator, and is the maximu m-likelihood estimate when the popu lation is normally distribu ted.

Properties of standard deviation
For constant c and random variables X and Y:

where

and

stand for variance and covariance, respectively.

Interpretation and application
A large standard deviation indicates that the data points are far from the mean and a small standard deviation indicates that they are clu stered closely arou nd the mean. For example, each of the three data sets {0, 0, 14, 14}, {0, 6, 8, 14} and {6, 6, 8, 8} has a mean of 7. T heir standard deviations are 7, 5, and 1, respectively. T he t hird set has a mu ch smaller standard deviation than the other two becau se its valu es are all close to 7. In a loose sense, the standard deviation tells u s how far from the mean the data points tend to be. It will have the same u nit s as the data points themselves. If, for instance, the data set {0, 6, 8, 14} represents the ages of fou r siblings in years, the standard deviation is 5 years.

7 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

As another example, the data set {1000, 1006, 1008, 1014} may represent the distances traveled by fou r athletes, measu red in met ers. It has a mean of 1007 meters, and a standard deviation of 5 meters. Standard deviation may serve as a measu re of u ncertainty. In physical science for example, the reported standard deviation of a grou p of repeated measu rements shou ld give the precision of those measu rements. When deciding whether measu rements agree with a theoretical prediction, the standard deviation of those measu rements is of cru cial importance: if the mean of the measu rements is too far away from the prediction (with the distance measu red in standard deviations), then the theory being tested probably needs to be revised. T his makes sense since they fall ou tside the range of valu es that cou ld reasonably be expected to occu r if the prediction were correct and t he standard deviation appropriately qu antified. See prediction interval.

Application examples
T he practical valu e of u nderstanding t he standard deviation of a set of valu es is in appreciating how mu ch variation there is from the "average" (mean). Weather As a simple example, consider average temperatu res for cities. While two cities may each have an average temperatu re of 15 °C, it's helpfu l to u nderstand that the range for cities near the coast is smaller than for cities inland, which clarifies that, while the average is similar, the chance for variation is greater inland than near the coast. So, an average of 15 occu rs for one cit y with highs of 25 °C and lows of 5 °C, and also occu rs for another city with highs of 18 and lows of 12. T he standard deviation allows u s to recognize that t he average for the city with the wider variation, and thu s a higher standard deviation, will not offer as reliable a prediction of temperatu re as the city with the smaller variation and lower standard deviation. Sports Another way of seeing it is to consider sports teams. In any set of categories, there will be teams that rate highly at some things and poorly at others. Chances are, the teams that lead in the standings will not show su ch disparity, bu t will perform well in most categories. T he lower the standard deviation of their ratings in each category, the more balanced and consistent they will tend to be. Whereas, teams with a higher standard deviation will be more u npredictable. For example, a team that is consistently bad in most

8 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

categories will have a low standard deviation. A team that is consistently good in most categories will also have a low standard deviation. However, a team with a high standard deviation might be the type of team that scores a lot (strong offense) bu t also concedes a lot (weak defense), or, vice versa, that might have a poor offense bu t compensates by being difficu lt to score on. T rying to predict which teams, on any given day, will win, may inclu de looking at the standard deviations of the variou s team "stats" ratings, in which anomalies can match strengths vs. weaknesses to attempt to u nderstand what factors may prevail as stronger indicators of eventu al scoring ou tcomes. In racing, a driver is timed on su ccessive laps. A driver with a low standard deviation of lap times is more consistent than a driver wit h a higher standard deviation. T his information can be u sed to help u nderstand where opportu nities might be fou nd to redu ce lap times. Finance In ﬁnance, standard deviation is a representation of the risk associated with a given secu rity (stocks, bonds, property, etc.), or the risk of a portfolio of secu rities (actively managed mu tu al fu nds, index mu tu al fu nds, or ET Fs). Risk is an important factor in determining how to eﬃcient ly manage a portfolio of investments becau se it det ermines the variation in retu rns on the asset and/or portfolio and gives investors a mathemat ical basis for investment decisions (known as mean-variance optimizat ion). T he overall concept of risk is t hat as it increases, the expected retu rn on the asset will increase as a resu lt of the risk premiu m earned – in other words, investors shou ld expect a higher retu rn on an investment when said investment carries a higher level of risk, or u ncertainty of that retu rn. When evalu ating investments, investors shou ld estimate both the expect ed retu rn and the u ncertainty of fu tu re retu rns. Standard deviation provides a qu antiﬁed estimate of the u ncertainty of fu tu re retu rns. For example, let's assu me an investor had to choose bet ween two stocks. Stock A over the last 20 years had an average retu rn of 10%, with a standard deviation of 20% and Stock B, over the same period, had average retu rns of 12%, bu t a higher standard deviation of 30%. On the basis of risk and retu rn, an investor may decide that Stock A is the safer choice, becau se Stock B's additional 2% point s of retu rn is not worth the additional 10% standard deviation (greater risk or u ncertainty of the expected retu rn). Stock B is likely to fall short of the initial investment (bu t also to exceed the initial investment) more often than Stock A u nder the same circu mstances, and is estimated to retu rn only 2% more on average. In this example, Stock A is expected to earn abou t 10%, plu s or minu s 20% (a range of 30% to -10%), abou t two-thirds of the fu tu re year retu rns. When considering more extreme
9 of 16 02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

possible retu rns or ou tcomes in fu tu re, an investor shou ld expect resu lts of u p to 10% plu s or minu s 90%, or a range from 100% to -80%, which inclu des ou tcomes for three standard deviations from the average retu rn (abou t 99.7% of probable retu rns). Calcu lating the average retu rn (or arit hmetic mean) of a secu rity over a given nu mber of periods will generate an expected retu rn on t he asset. For each period, su btracting the expected retu rn from the actu al retu rn resu lts in the variance. Squ are the variance in each period to find the effect of the resu lt on the overall risk of the asset. T he larger the variance in a period, the greater risk the secu rity carries. T aking the average of t he squ ared variances resu lts in the measu rement of overall u nits of risk associated with the asset. Finding the squ are root of this variance will resu lt in the standard deviation of the investment t ool in qu estion.

Geometric interpretation
T o gain some geometric insights, we will start with a popu lation of three 3 valu es, x1 , x2, x3. T his defines a point P = (x1 , x2, x3) in R . Consider the line L = {(r, r, r) : r in R}. T his is the "main diagonal" going throu gh the origin. If ou r three given valu es were all equ al, then the standard deviation wou ld be zero and P wou ld lie on L. So it is not u nreasonable to assu me t hat the standard deviation is related to the distance of P to L. And that is indeed the case. Moving orthogonally from P to the line L, one hits the point :

whose coordinates are the mean of the valu es we started ou t with. A little algebra shows that the distance between P and R (which is the same as the distance between P and the line L) is given by σ√3. An analogou s formu la (with 3 replaced by N) is also valid for a popu lation of N valu es; we then have N to work in R .

Chebyshev's inequality
An observation is rarely more than a few standard deviat ions away from the mean. Chebyshev's inequ ality entails t he following bou nds for all distribu tions for which the standard deviation is defined. At least 50% of the valu es are within √2 standard deviations from the mean. At least 75% of the valu es are within 2 standard deviations from the mean. At least 89% of the valu es are within 3 standard deviations from the mean.
10 of 16 02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

At least mean. At least mean. At least mean. At least mean.

94% of the valu es are within 4 standard deviations from the 96% of the valu es are within 5 standard deviations from the 97% of the valu es are within 6 standard deviations from the 98% of the valu es are within 7 standard deviations from the

And in general:
2 At least (1 − 1/k ) × 100% of the valu es are within k standard deviations from the mean.

Rules for normally distributed data
T he central limit theorem says that the distribu tion of a su m of many independent, identically distribu ted random variables tends towards the normal distribu tion. If a data distribu tion is approximately normal then abou t 68% of the valu es are within 1 standard deviation of the mean (mathematically, μ ± σ, where μ is the arithmetic mean), abou t 95% of the valu es are within two standard deviations (μ ± 2σ), and abou t 99.7% lie within 3 standard deviations (μ ± 3σ). T his is known as the 68-95-99.7 rule, or the empirical rule.

For variou s valu es of z, the percentage of valu es expected to lie in the symmetric conﬁdence interval (−zσ,zσ) are as follows:
zσ 1σ 1.645σ 90% percentage 68.27%

Dark blue is less than one standard deviation from the mean. For the normal distribution, this accounts for 68.27 % of the set; while two standard deviations from the mean (medium and dark blue) account for 95.45%; three standard deviations (light, medium, and dark blue) account for 99.73%; and four standard deviations account for 99.994%. The two points of the curve which are one standard deviation from the mean are also the inflection points.

11 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

1.960σ 2σ 2.576σ 3σ 4σ 5σ 6σ 7σ

95% 95.450% 99% 99.7300% 99.993666% 99.99994267% 99.9999998027% 99.9999999997440%

3.2906σ 99.9%

Relationship between standard deviation and mean
T he mean and the standard deviation of a set of data are u su ally reported together. In a certain sense, the standard deviation is a "natu ral" measu re of statistical dispersion if the center of the data is measu red abou t the mean. T his is becau se the standard deviation from the mean is smaller than from any other point. T he precise statement is the following: su ppose x1 , ..., xn are real nu mbers and define the fu nction:

Using calcu lu s, or simply by completing the squ are, it is possible to show that σ(r) has a u niqu e minimu m at the mean:

T he coefficient of variation of a sample is the ratio of the standard deviation to the mean. It is a dimensionless nu mber that can be u sed to compare the amou nt of variance between popu lations with different means. If we want to obtain the mean by sampling the distribu tion then the standard deviation of the mean is related to the standard deviation of the distribu tion by:

12 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

where N is the nu mber of samples u sed to sample the mean.

Rapid calculation methods
See also: Algorithms for calculating variance A slightly faster (significantly for ru nning standard deviat ion) way to compu t e the popu lation standard deviation is given by the following formu la (thou gh considerations mu st be made for rou nd-off error, arithmetic overflow, and arithmetic u nderflow conditions):

T he following two formu las are a u sefu l representation of ru nning (continu ou s) standard deviation. A set of three power su ms s0,1,2 are each compu ted over a set of N valu es of x, denoted as x k. Given the resu lts of these three ru nning su mations, one can u se σ at any time to compu te the current valu e of the ru nning standard deviation. T his crafty definition for sj allows u s to easily represent the two different phases (su mmation compu tation sj, and σ calcu lation). Note that s0 raises x to the zero power, 0 and since x is always 1, s0 evalu ates to N.

where the power su ms s0 , s1 , s2 are defined by

In a compu ter implementation, as the three sj su ms become large, we need to consider rou nd-off error, arithmetic overflow, and arit hmetic u nderflow. T o avoid this, we will periodically redu ce their absolu te valu es in a process reminiscent of normalizing a u nit vector. Since s1 is the su m of valu es and s2 is the su m of squ ares, we can estimate these valu es for a smaller valu e of N simply by dividing by ou r cu rrent N, and mu ltiplying by a well-selected
13 of 16 02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

smaller new-N. Ou r comparison with a u nit vector encou rages u s to consider selecting 1 as the valu e of new-N. However, this is a particu larly poor choice, as the accu racy of ou r continu ou s approximation was est ablished only for large N, and this wou ld cau se ou r next valu e to have as mu ch weight in the calcu lation as all previou s valu es. A more appropriate valu e of new-N is the maximu m valu e we can afford, su ch that we are su re we can renormalize back to new-N again before N again becomes large enou gh to introdu ce error (or catastrophe) as we add more valu es. Similarly for sample standard deviation:

Or from ru nning su ms:

T he above method can be very su sceptible to rou nding, u nderflow, and overflow errors, especially when the sample valu es are very close to the mean. It can actu ally give negative standard devation valu es, which shou ld be impossible given the definition. T his method is also given in a lot of textbooks. However, it shou ld not be u sed. Below is a better method for calcu lating ru nning su ms method with redu ced rou nding errors: A1 = x1

where A is the mean valu e. Q1 = 0

sample variance:

standard variance

14 of 16

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

For weighted distribu tion it is somewhat more complicated: T he mean is given by: A1 = x1

where wj are the weights. Q1 = 0

where n is the total nu mber of elements, and n' is the nu mber of elements with non-zero weights. T he above formu las become equ al to the more simple formu las given above if we take all weights equ al to 1.

Accu racy and precision Algorithms for calcu lating variance An inequ ality on location and scale parameters Chebyshev's inequ ality Confidence interval Cu mu lant Deviation (statistics) Geometric standard deviation Ku rtosis Mean absolu te error
15 of 16

Pooled standard deviation Raw score Root mean squ are Sample size Satu ration (color theory) Skewness Standard error Standard score Unbiased estimation of standard deviation Variance Volatility Yamartino method for calcu lating

Statistics portal

02/02/2009 10:16 AM

Standard deviation - Wikipedia, the free encyclopedia

http://en.wikipedia.org/wiki/Standard_deviation

Mean Median

standard deviation of wind direction

References
1. ^ Sir Francis Galton discovered the standard deviation (http://www.sciencetimeline.net/1866.htm)