EXST7005 Fall2010 07a Z distribution & Sample Means

EXST7005 Fall2010 07a Z distribution & Sample Means -...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Statistical Methods I (EXST 7005) Page 37 not, we can write this as two expressions, P(Z ≤ Z1) OR P(Z ≥ Z2). It is also possible that problems involving sections may not be symmetric, and may occur entirely in the positive tail, or negative tail. Some Examples P(|Z| ≥ Z0), where Z0 = 1.96. Since we are taking the absolute value of a randomly chosen Z value in either tail, that value may be either positive or negative and its absolute value may be greater than or equal to 1.96. P(|Z| ≥ Z0), where Z0 = 1.96. -4 -3 -2 -1 -3 -2 0 -1 1 2 3 4 P(|Z| ≥ Z0) = P(Z ≤ –Z0) + P(Z ≥ Z0), and since it is symmetric P(|Z| ≥ Z0) = 2*P(Z ≥ Z0) = 2(0.0250) = 0.050 P(|Z| ≤ Z0), where Z0 = 2.576. This problem is similar to the previous, but it describes the area in the middle, between two limits. -4 P(|Z| ≤ Z0) = 1 – P(|Z| ≥ Z0) = 1–2*P(Z ≥ Z0) = 1 – 2(0.0050) = 0.99 0 1 2 3 4 1 2 3 4 1 2 3 4 An asymmetric case P(–1.96 ≤ Z ≤ 2.576) = ? This is the area in the middle, the total minus the two tails. We already know these tails. P(–1.96 ≤ Z ≤ 2.576) = 1 – P(Z ≥ 1.96) – P(Z ≥ 2.576) = 1 – 0.0250 – 0.0050 = 0.97 -4 -3 -2 -1 0 Working the Z tables, backward & forward We have seen how to find a probability from a value of Z0. Now we need to be able to find a value of Z0 when a probability is known. Basically, we find the value of the probability in the body of our Z table, and determine the corresponding value of Z0. P(Z ≤ Z0) = 0.1587, find the value of Z0 -4 -3 -2 -1 0 This probability is a value less than 0.5, so it is a tail and can be solved directly from our tables. We only need to find 0.1587 in the table and determine the corresponding value of Z0. The value in the table occurs in the row corresponding to “1.0” and the column corresponding to “0.00”. Finally note that the randomly chosen Z was to be less than or equal to Z0, so we are in the lower tail. Z0 = –1.00 P(Z ≤ Z0) = 0.8413, find the value of Z0 This probability is a value greater than 0.5. To read it from our tables we must determine the corresponding tail. The tail would be given by 1– 0.8413 = 0.1587. So this is the same as the value we just looked up, it occurs in the row corresponding to “1.0” and the column -4 -3 -2 -1 0 1 2 3 4 James P. Geaghan Copyright 2010 Statistical Methods I (EXST 7005) Page 38 corresponding to “0.00”. Finally, note that the randomly chosen Z was to be less than or equal to Z0, but since the probability was larger than 0.5 we are in the upper tail. Z0 = +1.00 P(1 ≤ Z ≤ Z0) = 0.1, find the value of Z0 First note that the lower limit is 1.00, so we are working with the upper tail of the Z distribution. We can find the probability that Z is ≤ 1 as 1 – P(Z ≥ 1) = 1 – 0.1587 = 0.8413. The area of interest is stated to have an area of 0.1, so the upper tail would be what remains in the upper tail, 1 – 0.8413 – 0.1 = 0.0587. We find 0.0587 in our Z tables and note that it is between 1.56 and 1.57. I would accept either answer. The actual value, determined by the EXCEL “NORMDIST” function is Z0 = 1.565781531. -4 -3 -2 -1 0 1 2 3 4 Z 0 s o m e w h e re h e re 0 .1 0 .0 5 8 7 0.8 4 1 3 -4 -3 -2 -1 0 1 2 3 4 P(|Z| ≥ Z0) = 0.05, find the value of Z0 Note that the random value of Z is greater than Z0, so we are examining an area in the tails, where the random Z may be either positive or negative and its absolute value will exceed the value of Z0. Recall that our tables give only one tail, and this particular case has 0.05 in two tails, so we want to determine the value of Z0 for only one of those tails to match our tables. Each tail would have half of the 0.05 since it is a symmetric problem. So the area in the tail is 0.05/2 = 0.025. We can examine the tables and determine that for a probability of 0.025 the Z value is 1.96, so Z0 = 1.96. Some people like to write Z0 = ±1.96, 1.96 -1.96 but this isn't really necessary for this case. P(|Z| ≤ Z0) = 0.99, find the value of Z0 Note that the random value of Z is less than Z0, so we are examining an area in the middle of the distribution. Again, the randomly chosen Z may be either positive or negative. Since our tables give the tails of the distribution we need to determine the area in the tails. Since 0.99 occurs in the middle, the tails must have an area of 1 – 0.99 = 0.01. Half of this is in each tail, so the area in one tail would be 0.01/2 = 0.005. Given that the area in the tail is 0.005, the Z value is 2.576 (actually 2 .5 7 6 -2 .5 7 6 between 2.57 and 2.58; this is a value we have seen previously). The Z transformation of any normal distribution Most practical applications of the Z distribution will require that we take a real distribution {N(μ,σ2)} and convert it to a Z distribution {N(0,1)}, calculate some probability statement, and then convert the results back to the original distribution. So we work with two distributions James P. Geaghan Copyright 2010 Statistical Methods I (EXST 7005) Page 39 Y ~ N(μ,σ2) Z ~ N(0,1) In order to convert the observed Y distribution to the more workable Z distribution we need to transform the distribution. We use the Z transformation. Zi = (Yi – μ) / σ For example, suppose we have a distribution where Y ~ N(20,16) and we wish to determine P(Y ≤ 24). We convert the probability statement for the original distribution into a Z distribution using our transformation. 20 Transformation: Zi = (Yi – μ) / σ for N(20, 16) ⎛ (Yi − μ ) P(Y ≤ 24) = P ⎜ ⎝ ( σ ≤ 24 − 20) ⎞ 4⎟ ⎠ 24 0.8413 = P(Z ≤ 1) P(Z ≤ 1) = 1 – P(Z ≥ 1)=1 – 0.1587 = 0.8413 20 0 24 1 Original scale Z distribution scale So, P(Y ≤ 24) = 0.8413 Another example Using the same distribution Y ~ N(20,16) where we wish to determine P(Y ≥ 22). ⎛ (Y − μ ) ( 22 − 20) ⎞ = P(Z ≥ 0.5) P(Y ≥ 22) = P ⎜ i σ ≤ 4⎟ ⎝ ⎠ 0.30 8 5 20 0 22 0.5 From the table we determine that P(Z ≥ 0.5) = 0.3085 Another type of example Find Y0, where P(Y ≤ Y0) = 0.1515 for the same distribution, Y~N(20, 16). Again, using our transformation, ⎛ (Yi − μ ) P(Y ≤ Y0) = 0.1515 = P ⎜ ⎝ σ ≤ (Y0 − μ ) ⎞ σ ⎟= ⎠ Z = 1 .0 3 P(Z ≤ Z0) = 0.1515 Notice that the sign is ≤ and the probability small (less than 0.5), so we are in the lower half of the distribution. From the table the Z value is 1.03, so Z0 = –1.03. So we know that Z0 = –1.03, an area in the lower half of the distribution, but we don't know the value of Y0 yet and our original problem was to find P(Y ≤ Y0) = 0.1515. So we need to transform back to the Y scale. This back-transformation is a reversal of the Z transformation. Transformation: Zi = (Yi – μ) / σ To transformation back calculate: James P. Geaghan Copyright 2010 Y=15.88 Statistical Methods I (EXST 7005) Page 40 Yi = μ + Zσ = 20 + (–1.03)4 = 20 + (–4.12) = 20 – 4.12 = 15.88. The interpretation of this problem is that there is a probability of 0.1515 of randomly drawing a value from this distribution and getting a value that is less than 15.88. One last example Determine within what limits 50% of the distribution lies. We will use the same hypothetical distribution Y~N(20,16). We will assume that the objective here is to form symmetric limits as shown below. Otherwise, the problem has no answer. After all, half of the distribution (50%) lies above Z0 = 0, so the limits could be (0, ∞). Half the distribution also lies below Z0 = 0, so the limits could also be (–∞, 0). And there are an infinite number of other such limits in between. So, assuming symmetry, the problem becomes to determine the values of Y0 such that half of the distribution lies between –Z1 and +Z2, where Z1 = Y2 due to symmetry and we will refer to this value as Y0. Once we determine the values of Z that meet the probability limits we will transform the upper and lower value back to the Y distribution limits. 0.50 Determine: 0.25 0.25 P(|Z| ≤ Z0) = 0.5 Z0 -Z 0 Since our tables give area in the tails, we need to change this from P(|Z| ≤ Z0) = 0.5 to 12P(Z ≥ Z0) = 0.5 and find P(Z ≥ Z0) = (1–0.5)/2 or P(Z ≥ Z0) = 0.25. From our table: Z0 = 0.675 Transforming Z0 = 0.675 back to the Y scale gives, –Z0, Y1 = 20 – 0.675(4) = 20 – 2.7 = 17.3 Z0, Y2 = 20 + 0.675(4) = 20 + 2.7 = 22.7 The final probability statement is best given in the form, P(17.3 ≤ Y ≤ 22.7) = 0.50 Summary on use of Z tables Values of the relative cumulative frequency are given in the table. 0.50 0.25 0.25 -Z 0 =-0.675 Y 1 =17.3 Z 0=0.675 Y 2 =22.7 the table is one – sided the value given in the table is for the upper tail of the distribution The total area under the curve is 1.0 The distribution is symmetrical. e.g. for Z = –1 r.c.f. = 0.1587 (in the left tail) for Z = 1 r.c.f. = 0.1587 (in the right tail) We can transform from any normal distribution to the Z distribution by using the transformation Zi = (Yi – μ) / σ We can transform back from the Z distribution to any normal distribution by using the transformation Yi = μ + Zσ James P. Geaghan Copyright 2010 Statistical Methods I (EXST 7005) Page 41 Memorable values of Z (for 2 sided evaluations) μ ± 1σ = 0.68 μ ± 1.645σ = 0.90 μ ± 1.96σ = 0.95 μ ± 2.576σ = 0.99 A tip. Some tables give the one tailed probabilities at the top, some the two tailed probabilities. Some are cumulative from the lower end, some cumulative from the zero value (middle) up, and some give the cumulative area in the tails (like mine). If you know that 1.96 leaves 2.5% in one tail and 5% in two tails you can look for this value and figure out what kind of tables you have. Distribution of sample means Sample means are the basis for testing hypotheses about μ, the most common types of hypothesis tests. Before discussing hypothesis tests we will need some additional information about the nature of sample means. Imagine we are drawing samples from a population with the following characteristics. Population size = N Mean = μ Variance = σ2 The individual observations from this parent population are: Yi = Y1, Y2, Y3, ... , YN The set of samples of size n from the parent population form a derived population There are Nn possible samples of size n that can be drawn from a population of size N (sampling with replacement, which simulates a very large population). n For each sample we calculate a mean Yk = ∑Y i =1 ik n , where k = 1, 2, 3, ... , Nn The Derived Population of means of samples of size n The sample size = n Population size = Nn Mean = μ Y Variance = σ Y2 Derived population values Yk = Y1 , Y2 , Y3 , Y4 ..., YN n , Nn Mean of the derived population μY = ∑Y k =1 k Nn where k = 1, 2, 3, ... , Nn James P. Geaghan Copyright 2010 Statistical Methods I (EXST 7005) Page 42 Nn 2 Variance of the derived population σ Y = ∑ (Y k =1 k − μ )2 where k = 1, 2, 3, ... , Nn Nn Original Population Example of a Derived Population Parent Population: Yi = 0, 1, 2, 3 0.25 r.f. n = 2 and Nn = 42 = 16 where 0.00 Draw all possible samples of size 2 from the Parent 0 1 2 3 Population (sampling with replacement, so that values will occur more than once), and calculate a mean for each of the Nn samples. For this discrete uniform distribution the Mean = (Max + Min)/2 = (3-0)/2 = 1.5 and the variance = (Max–Min+1)2 /12 = (3–0+1)2/12 = 1.25 and the standard deviation = 1.1180. Sampling results for all possible means with replacement. Sample 0, 0 0, 1 0, 2 0, 3 1, 0 1, 1 1, 2 1, 3 2, 0 2, 1 2, 2 2, 3 3, 0 3, 1 3, 2 3, 3 Mean 0.0 0.5 1.0 1.5 0.5 1.0 1.5 2.0 1.0 1.5 2.0 2.5 1.5 2.0 2.5 3.0 Deviation from 1.5 Squared Deviation -1.5 -1 -0.5 0 -1 -0.5 0 0.5 -0.5 0 0.5 1 0 0.5 1 1.5 2.25 1 0.25 0 1 0.25 0 0.25 0.25 0 0.25 1 0 0.25 1 2.25 Distribution of the derived population of sample means Means 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Sum = Frequency 1 2 3 4 3 2 1 16 Relative Freq 0.0625 0.1250 0.1875 0.2500 0.1875 0.1250 0.0625 1 0.30 0.25 0.20 0.15 0.10 0.05 0.00 Derived Population 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Nn μY = ∑ Yk N n = 2416 = 1.5 k =1 Nn σ = ∑ (Yk − μ ) 2 Y k =1 2 N n ( 4(0.0) = 2 + 6(±0.5)2 + 4(±1.0)2 + 2(±1.5) 2 ) 16 = James P. Geaghan Copyright 2010 Statistical Methods I (EXST 7005) Page 43 ( 4(0.0) + 6(0.25) + 4(1.0) + 2(2.25) ) = 10 = 0.625 16 16 σ = 0.7906 Note that the histogram of the derived population shows that the population is shaped more like the normal distribution than the original population. Probability statement from the two distributions: Find P(1 ≤ Y ≤ 2) For the original, uniform population, P(1 ≤ Y ≤ 2) = 0.5000 For the derived population, P(1 ≤ Y ≤ 2) = 0.6250 THEOREM on the distribution of sample means Given a population with mean μ and variance σ2, if we draw all possible samples of size n (with replacement) from the population and calculate the mean, then the derived population of all possible sample means will have Mean: μY = μ Variance: 2 σY = σ n Standard error: 2 σY = σ n = σ n 2 Notice that the variance and standard deviation of the mean have “n” in the denominator. As a result, the variance of the derived population becomes smaller as the sample size increases, regardless of the value of the population variance. Central Limit Theorem As the sample size (n) increases, the distribution of sample means of all possible samples, of a given size from a given population, approaches a normal distribution if the variance is finite. If the base distribution is normal, then the means are normal regardless of n. Why is this important? (It is very important!) If we are more interested in the MEANS (and therefore the distribution of the means) than the original distribution, then normality is a more reasonable assumption. Often, perhaps even usually, we will be more interested in characteristics of the distribution, especially the mean, than in the distributions of the individuals. Since the mean is often the statistic of interest it is useful to know that it is possibly normally distributed regardless of the parent distribution. NOTES on the distribution of sample means Another property of sample means as n increases, σ Y2 and σ Y decrease. σ Y2 ≤ σ 2 σ Y2 < σ 2 for any n for any n > 1 James P. Geaghan Copyright 2010 Statistical Methods I (EXST 7005) Page 44 as n increases and σ Y becomes smaller, the distribution of the means gets closer to μ Y . (i.e. we get a better estimate). Some new terms Reliability (as a statistical concept) – the less scatter that occurs in a set of data, the more “reliable” the estimate. This term is also associated with the word “precision” in statistics where high reliability comes from low variance and high precision. This concept is important in estimation because it suggests that data is reproducible or repeatable. Accuracy (as a statistical concept) – this term refers to the lack of bias in the estimate, and not the smallness of the variance (e.g. reliability or precision). An accurate, or unbiased, estimated may have considerable scatter among the points, but on average the center of the distribution is neither overestimated or underestimated. This aspect of data is related to the validity of data. 2 Estimation of σY and σY In practice we cannot draw all possible samples. Recall that E(S2) = σ 2 2 so, SY = S 2 2 n is an estimate of σ Y where; S Y2 is an estimate of the variance of sample means of size n S2 is the estimate of the variance of observations 2 SY = S =S is called the standard error to distinguish it from the standard n n deviation it is also called the standard deviation of the means Notice that this is a division by “n “ for both populations and samples, not by “ n – 1” as with the calculation of variance for samples. S Y2 is a measure of the reliability of the sample means as an estimate of the true population mean. i.e. the smaller S Y2 , the more reliable Y as an estimate of μ Ways of increasing reliability Basically, anything that decreases our estimate of S Y2 makes our estimate more reliable. How do we decrease our estimate of S Y2 ? Increase the sample size; if n increases then S Y2 decreases. Decrease the variance; if our estimate of S 2 decreases then S Y2 decreases. This can sometimes be done: James P. Geaghan Copyright 2010 Statistical Methods I (EXST 7005) Page 45 • by refining our measurement techniques • by finding a more homogeneous population to measure (stratification) • by removing exogenous sources variance (blocking) Notation: The variance of a variable Y can be denoted S 2 or S Y2 . The subscript is only needed to clarify which variable is indicated, so it is often not needed and omitted. However, the variance of the means should be subscripted, S Y2 , to distinguish it from the variance of the observations. The Z transformation for a derived population We will use the Z transformation for two applications, individuals and means. Applications to individual observations – to make statements about individual members of the population, such as “What percentage of the individuals would be expected to have values greater than 17?” Zi = (Yi − μ ) σ Applications to means – to make statements about population means such as “What is the probability that the true population mean is greater than 17?” Zi = (Y − μ ) i Y σY Summary Most testing of hypotheses will concern tests of a derived population of means. The mean of the derived population of sample means is μ Y The Variance of the derived population of sample means is σ Y2 The CENTRAL LIMIT THEOREM is an important aspect of hypothesis testing because it states that sample means tend to be more nearly normally distributed than the parent population. We will often work with distributions that are not normally distributed, but the fact that we are often interested in the more normally distributed means instead of the original observations allows the use of parametric statistics. Reliability and accuracy are statistical concepts relating to variability and lack of bias, respectively. Mean: μY = μ Variance: 2 σY = σ n Standard error: 2 σY = σ n = σ n 2 James P. Geaghan Copyright 2010 ...
View Full Document

This note was uploaded on 12/29/2011 for the course EXST 7005 taught by Professor Geaghan,j during the Fall '08 term at LSU.

Ask a homework question - tutors are online