ࡱ>  gbjbjdd <_44wwwwwLL'''' LLLLLLL$uNQV2Lw"2Lww''[GLw'w' L LFDI's^HG&K]L0LnGfmQ^jmQLImQwI$2L2LLmQ4 =: Calculating Appropriate Sample Sizes Using Power Analysis Objective You're about to embark on a research project that will involve collecting data and performing tests of statistical inference, and you'd like to start your data collection. But as a savvy practitioner of statistics, you know how important is to estimate your sample size before you start collecting data, to ensure that you will be able to generate results that are statistically sound. There is nothing worse than being all ready to compute your results and generate your conclusions and then realizing... oh no, I don't have enough data for my results to be valid! (This happened to me when I was preparing my first dissertation proposal. Trust me, it can be not only unpleasant, but soul crushing.) For each of your research questions, you will perform one sample size calculation using R; then, select the largest number from this collection. That's the minimum sample size you should collect to ensure that the conclusions you draw from each research question are valid. You'll have to do some estimating (and make some educated guesses), but this is OK. Just be sure to articulate what assumptions you made when you were computing your appropriate sample size, and what trade-offs factored into your final decision. Feel free to consult the academic literature or do a pilot study to help support your estimates and guesses. Background Sample size calculation is an aspect of power analysis. If the population really is different than what you thought it was like when you set up your null hypothesis, then you want to be able to collect enough data to detect that difference! Similarly, sometimes you know you'll only have a small sample to begin with, so you can use power analysis to make sure you'll still be able to carry out a legitimate statistical study at a reasonable level of significance (. The power of a statistical test, then, answers the question: What's the probability that I will have enough data to determine that what I originally thought the population was like (as expressed by my null hypothesis) was incorrect? Clearly, having a zero percent chance of being able to use your sample to detect whether the population is unlike what you originally thought... would be bad. Similarly, having a 100% chance of being able to detect a difference would also probably be bad: your sample size would be comparatively large, and collecting more data is usually more costly (in both time and effort). When determining an appropriate sample size for your study, look for a power of at least 0.80, although higher is better; higher power is always associated with bigger sample sizes, though. The standard of 0.80 or higher just reflects that the community of researchers are usually comfortable with studies where the power is at least at this level. The smaller the difference between what you think the population is like (as described by the null hypothesis) and what the data tells you the population is like, the more data you'll need to actually detect that difference. This difference between your hypothesized mean and the sample mean, or the hypothesized proportion and the sample proportion, or the hypothesized distribution of values over a contingency table and the actual distribution of values in that contingency table, is called an effect size. For example, if you were trying to determine whether there was a difference between the average age of freshmen and the average age of seniors at your university, it wouldn't require such a large sample size because the effect size is about three years in age. However, if you were trying to determine whether there was a difference between the average age of freshmen and the average age of sophomores at your university, it would require a much larger sample size because the effect size is less than one year in age. If there is a difference, you will need more data in your sample to know for sure. Effect size is represented in R as how many standard deviations the real estimate is away from the hypothesized estimate. Closer to zero, the effect size is small; around 0.5, the effect size becomes more significant. An effect size of 0.8 to 1 is considered large, because you're saying that the true difference between the hypothesized value of your population parameter and the sample statistic from your data is approaching one standard deviation. A value of 0.1 or 0.2 is a small effect size. (Cohen, 1988) Your job is to estimate what you think the effect size is before you perform the computation of sample size. If you really have no way at all of knowing what the effect size is (and this is actually a pretty common dilemma), just use 0.5 as recommended by Bausell & Li (2002). To increase or improve your power to detect an effect, do one or more of these: Increase your sample size. The more items you have in your sample, the better you will have captured a snapshot of the variation throughout the population... as long as your random sample is also a representative sample. Increase your level of significance, (, which will make your test less stringent. Increase the effect size. Of course, this is a characteristic of your data and your a priori knowledge about the population... so this one may be really hard to change. Use covariates or blocking variables. (I won't explain this in detail; just know that if you're designing an experiment with treatment and control groups, and you need to improve your experimental design to increase the power of your statistical test, you should look into this.) Now that you understand power and effect size, it will be much easier to get a sense of what Type I and Type II Errors are and what they really mean. You may have heard that Type I Error, (, reflects how likely you are to reject the null hypothesis when you shouldn't have done it. You know how sampling error means that on any given day, you might get a sample that's representative of the population, or you might get a sample that estimates your mean or proportion too high or too low? Type I Error asks (and note that all of these questions ask exactly the same thing, just in different ways): How willing am I to reject the null hypothesis when in fact, it actually pretty accurately represents what's going on with the population? How willing am I to get a false positive, where I detected an effect but no effect actually exists? What's the probability of incorrectly rejecting the null hypothesis? This probability, the Type I Error, is the level of significance (. If you choose an ( of 0.05, that means you are willing to be wrong like this 1 out of every 20 times (1/20 = 0.05) you collect a sample and run the test of inference. If you choose an ( of 0.01, that means you are willing to be wrong like this 1 out of every 100 times (1/100 = 0.01) you collect a sample and run the test of inference -- making this selection of ( a more stringent test. On the other hand, if you choose an ( of 0.10, that means you are willing to be wrong like this 1 out of every 10 times (1/10 = 0.10) you collect a sample and run the test of inference -- making this selection of ( a much less stringent test. Type I Error and Type II Error have to be balanced depending upon what your goals are in designing your study. Here is how they are related: What's really going on with the populationH0 is TrueH0 is FalseThe decision you make as a result of your statistical test:Reject H0Type I Error ( FALSE POSITIVESAccurate Results! You rejected H0 and you were supposed to, because your data showed that the population was different than what you originally thoughtFail to Reject H0Accurate Results! You didn't reject H0 because it was an accurate description of the populationType II Error ( FALSE NEGATIVES Power, 1 - (, is related to the Type II Error... it is: The probability that you DON'T get a FALSE NEGATIVE The probability that you DO detect an effect that's REAL Process So now it's time to compute your appropriate sample sizes. For each of your research questions, you should have already selected the appropriate methodology for statistical inference that you'll use to draw your conclusions. The inferential tests that are covered in this chapter are: One sample t-test Two sample t-test Paired t-test One-proportion z-test Two proportion z-test Chi-Square Test of Independence Analysis of Variance (ANOVA) Before we get started, here is a summary of some sample size calculation commands you can run in R. All of them are provided by the pwr package, except the last one, which is in the base R installation. Be sure to install the pwr package first, then load it into active memory using the library command, before you move on. R CommandStatistical Methodologypwr.t.testOne sample, two sample, and paired t-tests; also requires you to specify whether the alternative hypothesis will be one tailed or twopwr.t2n.testTwo sample t-test where the sizes of the sample from each of the two groups is differentpwr.p.testOne proportion z-testpwr.2p.testTwo proportion z-testpwr.2p2n.testTwo proportion z-test where the sizes of the sample from each of the two groups is differentpwr.chisq.testChi-square Test of Independencepwr.anova.testAnalysis of Variance (ANOVA)power.t.testAnother way to perform power analysis for one sample, two sample, and paired t-tests; here, the advantage is that there is an easy method to plot Type I & Type II Errors & power vs. effectCalculating Sample Sizes for Tests Involving Means (T-Tests) This section covers sample size calculations for the one sample, two sample, and paired t-tests. You are also required to specify whether your alternative hypothesis will be one tailed or two, so be sure you have defined your H0 and Ha prior to starting your calculations. The pwr.t.test command takes five arguments, which means you can use it to compute power and effect size in addition to just the sample size (if you want). So the one sample t-test, we can use pwr.t.test like this to compute required sample size: > pwr.t.test(n=NULL,sig.level=0.05,power=0.8,d=0.3,type="one.sample") One-sample t test power calculation n = 89.14936 d = 0.3 sig.level = 0.05 power = 0.8 alternative = two.sided > pwr.t.test(n=NULL,sig.level=0.05,power=0.8,d=0.3,type="one.sample", alternative="greater") One-sample t test power calculation n = 70.06793 d = 0.3 sig.level = 0.05 power = 0.8 alternative = greater To get the sample size, we use the n=NULL argument to pwr.t.test. As expected, the one-tailed test below requires a smaller sample size than the two-tailed test above. And always round your n's up!! We can't sample an extra 0.14936 person for the first test above.. we have to sample the entire person. So our correct sample size should be 90 for that test, and 71 for the test below. We can use the same command to determine sample sizes for the two sample t-test: > pwr.t.test(n=NULL,sig.level=0.05,power=0.8,d=0.3,type="two.sample", alternative="greater") Two-sample t test power calculation n = 138.0715 d = 0.3 sig.level = 0.05 power = 0.8 alternative = greater NOTE: n is number in *each* group And the paired t-test: > pwr.t.test(n=NULL,sig.level=0.05,power=0.8,d=0.3,type="paired", alternative="greater") Paired t test power calculation n = 70.06793 d = 0.3 sig.level = 0.05 power = 0.8 alternative = greater NOTE: n is number of *pairs* Observe that you can calculate any of the values if you know all of the other values. For example, if you know you can only get 28 pairs for your paired t-test, you can first see what the power would be if you kept everything else the same, and then you can check and see what would happen if the effect size were just a little bigger (and thus easier to detect with a smaller sample): > pwr.t.test(n=28,power=NULL,sig.level=0.05,d=0.3,type="paired", alternative="greater") Paired t test power calculation n = 28 d = 0.3 sig.level = 0.05 power = 0.4612366 alternative = greater NOTE: n is number of *pairs* > pwr.t.test(n=28,power=0.8,sig.level=0.05,d=NULL,type="paired", alternative="greater") Paired t test power calculation n = 28 d = 0.4821407 sig.level = 0.05 power = 0.8 alternative = greater NOTE: n is number of *pairs* In the first example, we used power=NULL to tell R that we wanted to compute a power value, given that we knew the number of pairs n, the estimated effect size d, the significance level of 0.05, and that we are using the "greater than" form of the alternative hypothesis. But a power of 0.46 is really not good, so we'll have to change something else about our study. If we force a power of 0.8, and instead use d=NULL to get R to compute the effect size, we find that using the 28 pairs of subjects we have available, we can detect an effect size that's about half a standard deviation from what we hypothesized at a power of 0.8 and a level of significance of 0.05. That's not so bad. You can also access the sample size n directly, if that's all you're interested in, like this: > pwr.t.test(n=NULL,sig.level=0.05,power=0.8,d=0.3,type="paired", alternative="greater")$n [1] 70.06793 Here is a summary of all the arguments you can pass to pwr.t.test to perform your sample size calculations and power analysis for tests of means: Argument to pwr.t.testWhat it meansnYour sample size! Set to NULL if you want to compute it.powerSets the desired power level. Best to set it at 0.80 or above! But beware: the higher the desired power, the bigger your required sample size will be.sig.levelSets the level of significance (. Typically this will be 0.1, 0.05, or 0.01 depending upon how stringent you want your test to be (the smaller numbers correspond to more stringent tests, like you might use for high-cost or high-risk scenarios).dThis is the effect size. A reasonable heuristic is to choose 0.1 for a small effect size, 0.3 for a medium effect size, and 0.5 for a large effect size.type=c(" ")Specify which t-test you are using: the one sample t-test ("one.sample"), the two sample t-test ("two.sample"), or the paired t-test ("paired")? Put that word inside the quotes.alternative=c(" ")Specify which form of the alternative hypothesis you'll be using... the one with the < sign ("less")? The one with the > sign ("greater")? Or the two-tailed test with the ( sign ("two.sided")? Put that word inside the quotes. Calculating Sample Sizes for Tests Involving Proportions (Z-Tests) This section covers sample size calculations for the one proportion and two proportion z-tests. You are also required to specify whether your alternative hypothesis will be one tailed or two, so be sure you have defined your H0 and Ha prior to starting your calculations. The pwr.p.test and pwr.2p.test commands take five arguments, which means you can use it to compute power and effect size in addition to just the sample size (if you want). So for the one proportion z-test, we can use pwr.p.test like this to compute required sample size: > pwr.p.test(h=0.2,n=NULL,power=0.8,sig.level=0.05, alternative="two.sided") proportion power calculation for binomial distribution (arcsine transformation) h = 0.2 n = 196.2215 sig.level = 0.05 power = 0.8 alternative = two.sided You can access the sample size directly here as well: > pwr.p.test(h=0.2,n=NULL,power=0.8,sig.level=0.05, alternative="two.sided")$n [1] 196.2215 And for the two proportion z-test, we can use pwr.2p.test like this to compute required sample size (or access it directly using the $n variable notation at the end): > pwr.2p.test(h=0.2,n=NULL,power=0.8,sig.level=0.05, alternative="two.sided") Difference of proportion power calculation for binomial distribution (arcsine transformation) h = 0.2 n = 392.443 sig.level = 0.05 power = 0.8 alternative = two.sided NOTE: same sample sizes > pwr.2p.test(h=0.2,n=NULL,power=0.8,sig.level=0.05, alternative="two.sided")$n [1] 392.443 Here are summaries of all the arguments you can pass to pwr.p.test and pwr.2p.test to perform your sample size calculations and power analysis for tests of means: Argument to pwr.p.testWhat it meansnYour sample size! Set to NULL if you want to compute it.powerSets the desired power level. Best to set it at 0.80 or above! But beware: the higher the desired power, the bigger your required sample size will be.sig.levelSets the level of significance (. Typically this will be 0.1, 0.05, or 0.01 depending upon how stringent you want your test to be (the smaller numbers correspond to more stringent tests, like you might use for high-cost or high-risk scenarios).hThis is the effect size. A reasonable heuristic is to choose 0.2 for a small effect size, 0.5 for a medium effect size, and 0.8 for a large effect size.alternative=c(" ")Specify which form of the alternative hypothesis you'll be using... the one with the < sign ("less")? The one with the > sign ("greater")? Or the two-tailed test with the ( sign ("two.sided")? Put that word inside the quotes. Argument to pwr.2p.testWhat it meansnYour sample size! Set to NULL if you want to compute it.powerSets the desired power level. Best to set it at 0.80 or above! But beware: the higher the desired power, the bigger your required sample size will be.sig.levelSets the level of significance (. Typically this will be 0.1, 0.05, or 0.01 depending upon how stringent you want your test to be (the smaller numbers correspond to more stringent tests, like you might use for high-cost or high-risk scenarios).hThis is the effect size. A reasonable heuristic is to choose 0.2 for a small effect size, 0.5 for a medium effect size, and 0.8 for a large effect size.alternative=c(" ")Specify which form of the alternative hypothesis you'll be using... the one with the < sign ("less")? The one with the > sign ("greater")? Or the two-tailed test with the ( sign ("two.sided")? Put that word inside the quotes.Calculating Sample Sizes for the Chi-Square Test of Independence This section covers sample size calculations for the Chi-Square Test of Independence, which is performed on a contingency table created by tallying up observations that fall in each of two categories. The purpose of this test is to determine whether the two categorical variables are independent or not (that there is some kind of dependency within the data; you won't be able to tell what, specifically, without further experimentation). The pwr.chisq.test command takes five arguments, and like the other tests, you can find out the value of any of them with the remaining values. For a 3x3 contingency table where you expect a moderate effect size, you would do this: > pwr.chisq.test(w=0.3,N=NULL,df=4,sig.level=0.05,power=0.8) Chi squared power calculation w = 0.3 N = 132.6143 df = 4 sig.level = 0.05 power = 0.8 NOTE: N is the number of observations Argument to pwr.chisq.testWhat it meansNYour sample size! Set to NULL if you want to compute it. Note: it's a CAPITAL N!powerSets the desired power level. Best to set it at 0.80 or above! But beware: the higher the desired power, the bigger your required sample size will be.sig.levelSets the level of significance (. Typically this will be 0.1, 0.05, or 0.01 depending upon how stringent you want your test to be.wThis is the effect size. A reasonable heuristic is to choose 0.1 for a small effect size, 0.3 for a medium effect size, and 0.5 for a large effect size.dfDegrees of freedom; calculate by taking one less than the number of rows in the table, and one less than the number of columns in the table... then multiply them together.Calculating Sample Sizes for One-Way Analysis of Variance (ANOVA) This section covers sample size calculations for the one-way ANOVA, which tests for equivalence between several group means, and aims to determine whether one of these means is not like the others. The pwr.anova.test command takes five arguments, and like the other tests, you can find out the value of any of them with the remaining values. For an ANOVA test with four groups, you would do this: > pwr.anova.test(k=4,f=0.3,sig.level=0.05,power=0.8) Balanced one-way analysis of variance power calculation k = 4 n = 31.27917 f = 0.3 sig.level = 0.05 power = 0.8 NOTE: n is number in each group Here are summaries of all the arguments you can pass to pwr.chisq.test: Argument to pwr.anova.testWhat it meansnYour sample size, in terms of number of observations per group! Set to NULL if you want to compute it. powerSets the desired power level. Best to set it at 0.80 or above! But beware: the higher the desired power, the bigger your required sample size will be.sig.levelSets the level of significance (. Typically this will be 0.1, 0.05, or 0.01 depending upon how stringent you want your test to be.fThis is the effect size. A reasonable heuristic is to choose 0.1 for a small effect size, 0.3 for a medium effect size, and 0.5 for a large effect size.kNumber of groups (Check the null hypothesis here... how many ('s do you see in it? That's your number of groups.)Plotting Power Curves One of the things I like to do when I am exploring my tradeoffs between Type I Error (that is, level of significance (, which I can control) and Type II Error (which I can adjust based on what power I choose for my test) is to plot the various sample sizes I will need to achieve a power of at least 0.80 for a range of ( values. With sample size n on the horizontal x-axis, and power on the vertical y-axis, these tradeoffs become easier to see. Here is some code to plot a power curve. Note that you will need a NEW power curve for each statistical test that you plan to do, and you'll have to change the effect size and the range of sample sizes you want to consider (based on your particular study): # Set the effect size and range of n's to consider in your plot effect.size <- 0.3 n <- seq(10,400,10) # Change these to reflect the TYPE OF STATISTICAL TEST you are doing: pwr.05 <- pwr.t.test(d=effect.size,n=n,sig.level=.05,type="two.sample", alternative="greater") pwr.01 <- pwr.t.test(d=effect.size,n=n,sig.level=.01,type="two.sample", alternative="greater") pwr.1 <- pwr.t.test(d=effect.size,n=n,sig.level=.1,type="two.sample", alternative="greater") # Now Plot the Power Curve plot(pwr.05$n,pwr.05$power,type="l",xlab="Sample Size (n)",ylab="Power",col="black") lines(pwr.1$n,pwr.1$power,type="l",xlab="Sample Size (n)",ylab="Power",col="blue") lines(pwr.01$n,pwr.01$power,type="l",xlab="Sample Size (n)",ylab="Power",col="red") abline(h=0.8) grid(nx=25,ny=25) title("Power Curve for Sample Size Estimation") legend("bottomright", title="Significance Level", c("sig.level = 0.01","sig.level=0.05","sig.level=0.1"), fill=c("red","black","blue")) The plot below shows a prediction of power by sample size for least stringent (blue) to most stringent (red) tests.  Conclusions What sample sizes did you determine for each of your research questions? Now pick the largest one of all of those calculations you made. Choose that as your appropriate sample size, then move forward and make great research. (Props to Neil Gaiman on that ending statement.) Other Resources:  HYPERLINK "http://en.wikipedia.org/wiki/Statistical_power" http://en.wikipedia.org/wiki/Statistical_power  HYPERLINK "http://en.wikipedia.org/wiki/Type_I_and_type_II_errors" http://en.wikipedia.org/wiki/Type_I_and_type_II_errors  HYPERLINK "http://osc.centerforopenscience.org/2013/11/03/Increasing-statistical-power/" http://osc.centerforopenscience.org/2013/11/03/Increasing-statistical-power/ Bausell, R. B. and Li, Y. (2002). Power Analysis for Experimental Research: A Practical Guide for the Biological, Medical and Social Sciences. Cambridge University Press, New York, New York.  HYPERLINK "http://www.r-bloggers.com/power-and-sample-size-analysis-z-test/" http://www.r-bloggers.com/power-and-sample-size-analysis-z-test/ $:;<EFrT Z @ |    % & T \ o } 6 W w x y { hPhrhrhZNh); hP6hDW h3h3 h36h3h36hP hP6hPh3hKNhih65hihi5h%s>h$*5h%s>h%s>5 h-v5CJ hKN5CJ !jh%s>5CJ UmHnHu.<F x y WAr! " $a$gd-v $ & Fa$gda $ & Fa$gdP $^a$gd{$a$gda u(TUW[`v}6?@A+/ Xx pº²~yy h6hVEhVE5hV7 h hI%hrh-v h 6h  h})6 h{h})h\rLh{h{5h{h{6 h{h{ h{h);h{h);5h{h);h);5h); jahVEh hVE6hVEh}).L !&? ؼ{vv h\rL6 jah-vh\rLh-v5h\rLh-vhj=hNhN5hp h-5 h);5 h);6 jah);h);h);h);5hPhP5 hPhP hP6hP hhhrhr6hhhhrhh5- ! " 2 !+!@!!!!!R"S"##B#C###$<$D$$$$$$$$$$$$%,%-%.%;%<%=%M%^%%%%% &g&hAhd5CJhAh)T6CJ jahAhdCJhAhdCJhAhd5H*hAhd6hAhd5 h,>h,> h,>6h,> jah\rLhX3h\rL5h\rLh\rL6h\rLh\rL5h\rL h\rLh-v3" !V!$$$$$$$$$Ifa$gdAl $a$gd-v $a$gd\rL $ & Fa$gd\rL $$$$$$qYYYY$$Ifa$gdAl kd$$IflFx\ t06    44 laytA$$$%-$$Ifa$gdAl kd$$Ifl\xj\  t(0644 lap(ytA$%.%;%=%M%%$$Ifa$gdAl %%%%6$$Ifa$gdAl kd$$Ifl4\xj\  t(0644 lapytA%Y&g&i&y&$$Ifa$gdAl g&h&i&y&z&{&&&&&&&&&&&&&&&#'$'+','g'o'x'''H(I((((i)l)m)v))ʾʪ|soh2(hUhUCJhUhUCJOJQJ^JhU hPhPhChPhDWhih65 hP5 hdhdhhh,>hd jbhd5 hd5 h,>5 h\rL5hAhd5hAh)T6CJhAhdCJ jbhAhdCJ&y&z&{&&6, $^a$gdd $a$gd-vkd$$Ifl4\xj\   t(0644 lapytA&&&#'$','I([(m({(((((()*3*K*$d$Ifa$gdF9$a$gdP $^a$gdP $ & Fa$gdP$a$gda $a$gd-v $x^a$gdh $^a$gdd))))))* * *(*)*2*3*J*L*V*W*******$+/+5+6+7+9+C+E+O+P+e+g+r+s+++++,,$,ĸĸĸċċ{ohPCJOJQJ^Jh=CJOJQJ^Jh)h)CJOJQJ^Jh=h h CJOJQJ^JhuhPOJQJ^JhUCJOJQJ^JhPhUhUCJhUhUCJOJQJ^JhUh2(hUh2(CJhUh2(CJOJQJ^J+K*L*W**n`d$IfgdU$d$Ifa$gdF9kd$$Ifl  /0 \"     t0  644 laytF9***D+pbd$Ifgd=$d$Ifa$gd }kd$$Ifl  0 \"   t0  644 laytF9D+E+P+f+pbd$IfgdF9$d$Ifa$gdU}kd7$$Ifl  0 \" t0  644 laytF9f+g+s++pbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0 \" t0  644 laytF9++++pbd$Ifgd)$d$Ifa$gdF9}kdg$$Ifl  0 \" t0  644 laytF9++,%,pbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0 \" t0  644 laytF9%,&,5,R,pbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0 \" t0  644 laytF9$,&,4,5,Q,S,`,,, - ----Z-[->.?.E.F.k.l.p.z.{.../-/7/8/b/c/0ǿxok\h0h0CJOJQJ^Jh#Kh_ThX3CJh_ThX3CJOJQJ^JhX3 hDW6h_ThDWCJh_ThDWCJOJQJ^JhDW hq;hGqh; h)H*h)hGqh65 hH$5hXDhwhwCJOJQJ^Jh=hPCJOJQJ^Jh=CJOJQJ^JhP!R,S,`,-pbd$IfgdXD$d$Ifa$gdF9}kd/$$Ifl  0 \" t0  644 laytF9--[-l.c//////00yyqggggggg $a$gd0$a$gd#K$a$gda}kd$$Ifl  0 \" t0  644 laytF9 030O0P0000000 1!171Q1R12$3j33333333 4%4&4$a$gd#K $a$gd000Q1R1u1{1|111$3i3j3H4`444v5y55566 77;7<788R8S8(9G9R9[9]9999999::::;ݫ}th>h>CJh>h>CJOJQJ^Jh>h>hN-CJOJQJ^JhN-CJOJQJ^JhN-hN-CJOJQJ^J hN-6hN-h0h0CJ h0h0CJOJPJQJ^Jh0hX3h0h0CJOJQJ^Jh0CJOJQJ^J-&4H4I4`44444445)5?5Y5Z5w5x5y56<7S7T7z7{77$a$gdN- $a$gdN-$a$gd0$a$gd#K $a$gd077777788S8j8k8888888 9 9(9)9;8<z<<<< $a$gd>$a$gd0 $a$gdN-; <7<8<y<z<<<<<<<<3=4=@=J=S=X=Z=[=\=u=y=z======2>4>=>>>]>^>2?4?5?6????¹yun jahX3h_Th_ThX3CJhuh#KOJQJ^JhX3CJOJQJ^Jh#Kh_ThX3CJOJQJ^JhX3 h>h>hAhh>CJhAhh>CJOJQJ^Jh0CJOJQJ^Jh>CJOJQJ^Jh>h>CJOJQJ^Jh>hN-*<<4=K=Y=Z=\=K:$d$Ifa$gd kd_ $$Ifl  /02 \  *   t0  644 layt_T$d$Ifa$gdX3$d$Ifa$gd_T$a$gd> $a$gd0\====3>sbTd$Ifgd $d$Ifa$gd }kd $$Ifl  02 \ *  t0  644 layt_Td$Ifgd_T3>4>>>3?pbd$Ifgd $d$Ifa$gd }kd $$Ifl  02 \ * t0  644 layt_T3?4?6??pbd$Ifgd $d$Ifa$gd }kdG $$Ifl  02 \ * t0  644 layt_T???@pbd$Ifgd $d$Ifa$gd }kd $$Ifl  02 \ * t0  644 layt_T??@"@>@H@c@i@@@@@AA"A)AMANAVA_AAAAAAABBBBBBBBBBBBBBBB½­­©šrrih_ThRCJh_ThRCJOJQJ^JhRh_ThI@CJhRCJOJQJ^Jh_ThI@CJOJQJ^JhAhh; hI@H*hGqhH$5 hH$5hI@h#K jhX3hX3CJOJQJ^Jh hX3hX3CJOJQJ^JhX3h CJOJQJ^J)@@@Apbd$Ifgd $d$Ifa$gd }kdw $$Ifl  02 \ * t0  644 layt_TAAAAACD7DDDDyqqi_____ $a$gdAh$a$gdI@$a$gdH$$a$gd#K}kd $$Ifl  02 \ * t0  644 layt_T BBCCC CCCCCCCCCCCDDE:EmEnEEEEEEE>FrFsFGGGGH$H%H)H-H.H4H5HHＰԡＰthXFCCJOJQJ^JhAhhAhCJhAhhAhCJOJQJ^Jh_ThAhCJh_ThAhCJOJQJ^JhAhCJOJQJ^JhAhhAhCJOJQJ^Jh_ThI@CJhAhCJOJQJ^Jh_ThI@CJOJQJ^JhAh hI@6hI@hR+DDDDEE:EnEEEE>FsFFFFFG"G9GOGkGlGGGGG$a$gdAh$a$gdI@ $a$gdAhGGGHHHHHCkd $$Ifl  /0 \     t0  644 laytI@$d$Ifa$gdF9$d$Ifa$gdI@$a$gdAh$a$gdI@ $a$gdAhHHHHHHHHHHHHHHHIIIIIJJJJJJJKK"K5KKKKKKKKKLLL%L)L+L0L@LALh_ThAhCJOJQJ^Jh& jhI@hX3hI@CJOJQJ^JhAhhAhCJOJQJ^J jahI@h_ThI@CJhuhI@OJQJ^JhI@CJOJQJ^Jh_ThI@CJOJQJ^JhI@ h>hAh/HHHHIsb$d$Ifa$gdF9}kdQ$$Ifl  0 \   t0  644 laytI@d$IfgdF9IIIJpbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0 \  t0  644 laytI@JJJ!Kpbd$IfgdAh$d$Ifa$gdF9}kd$$Ifl  0 \  t0  644 laytI@!K"K5KLLpbbd$IfgdF9$d$Ifa$gdF9}kd'$$Ifl  0 \  t0  644 laytI@LL1L?Lpp$d$Ifa$gdF9}kd$$Ifl  0 \  t0  644 laytI@?L@LBL{Ln`d$IfgdF9$d$Ifa$gdF9kdW$$Ifl  /0 \     t0  644 laytF9ALBL[L_L`L|LLLM#M$MCMDMNNNNN'O+OIOPOtOuO}OOOOOP QQQQ+Q/QQQQQ視uiuhCJOJQJ^Jh_ThCJOJQJ^J h6h&h6h&h&6h&hhGqhH$5 hH$5 jhAhhX3hAhCJOJQJ^J jahAhhAhCJOJQJ^Jh_ThAhCJh_ThAhCJOJQJ^JhAhhuhAhOJQJ^J'{L|LLMpbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0 \   t0  644 laytF9MM$MNpbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0 \  t0  644 laytF9NNNNpbd$IfgdF9$d$Ifa$gdF9}kd?$$Ifl  0 \  t0  644 laytF9NNNOpbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0 \  t0  644 laytF9OOORRRRRS S4SKSyqgggggggg $a$gd&$a$gdIA$a$gdH$}kdo$$Ifl  0 \  t0  644 laytF9 QQQQ RRRRRSSSSSSSSSSSSSSSTTT T TTTTTT0U1U3U4U5UUUU}VVV漰椗漎気気気z hF95 jahIAh&h&5h_ThIACJhuhIAOJQJ^Jh&CJOJQJ^JhIACJOJQJ^Jh_ThIACJOJQJ^Jh&CJOJQJ^Jh&h&CJOJQJ^JhIAh& h6hh_ThCJ,KSaSbSSSSSSSSkd$$Ifl  /0N \     t0  644 laytIA$d$Ifa$gdF9$d$Ifa$gdIA $a$gd&STT TTsb$d$Ifa$gdF9}kd$$Ifl  0N \   t0  644 laytIAd$IfgdF9TTT2Upbd$Ifgd&$d$Ifa$gdF9}kdW$$Ifl  0N \  t0  644 laytIA2U3U5UUpbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0N \  t0  644 laytIAUUU~Vpbd$Ifgd&$d$Ifa$gdF9}kd$$Ifl  0N \  t0  644 laytIA~VVVNXXXXXXXYY5Yyyooooooooo $a$gdF9$a$gdF9}kd$$Ifl  0N \  t0  644 laytIA VV\WWWWWWWWWWW9XhMJhMJCJOJQJ^JhAhhMJCJOJQJ^JhF9CJOJQJ^JhF9hF9CJOJQJ^J hF9hF9hMJh_ThF9CJhF9CJOJQJ^Jh_ThF9CJOJQJ^J hF96hF9hGqhF95*5Y6YVYWYYYYYYY\kd$$Ifl  /0N \     t0  644 laytF9$d$Ifa$gdF9$a$gdMJ $a$gdF9 Y4Z5Z;ZZsb$d$Ifa$gdF9}kda$$Ifl  0N \   t0  644 laytF9d$IfgdF9ZZZ`[pbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0N \  t0  644 laytF9ZZZZa[b[c[[[<\=\r\\\\\ ]]4]:]]]]]]F^G^p^G_H_=`>`````Wa_abaoaaaaaaaaaaabbbbĽĵĵıĽıĔhiCJOJQJ^JhihiCJOJQJ^Jhi hi5 h)T6h[Fh)Th)T6 jah)Th)ThGqh)T5 h)T5 jmhF9hF9CJOJQJ^J jahF9hF9huhF9OJQJ^J5`[a[c[[pbd$IfgdF9$d$Ifa$gdF9}kd$$Ifl  0N \  t0  644 laytF9[[[q\pbd$IfgdF9$d$Ifa$gdF9}kd7$$Ifl  0N \  t0  644 laytF9q\r\\G^H______>`U```yyyrrrrrrrrrgdi$a$gd)T}kd$$Ifl  0N \  t0  644 laytF9 ``aa-aaa)b7bIbybbbccxczccdddee $ & Fa$gdP $ & Fa$gda$a$gda$a$gdigdib bb#b&b6b7bHbIbbbbbbbcwcxcyczcccccdRdedddddddddddƿ~vnj_njnhPUhPjhPUh##h65 hi5 hch6h}hH~hH~6hH~ hc6hchihi5 hQ9R5 hi5jghi5U hihihihihACJOJQJ^JhACJOJQJ^JhiCJOJQJ^JhihiCJOJQJ^J$deeee^e_e`eeeeeeeeAfBfCfgggUgVgWggggؼԭhbh,hb0Jjqh,Uh,jh,U hrhrh[hrh-v0JjdphrUhrjhrUjmohPUhPjhPUhPhP0JeCfgg $ & Fa$gd-v:&P 1h/R :pd0* 4!"#$% $$If!vh555#v#v:V l t0655/ /  / ytA$$If!vh5555#v:V l  t(065/ / / p(ytA$$If!vh5555#v:V l4  t(06+,,5/ pytA$$If!vh5555#v:V l4  t(06+,5/ pytA$$If!vh5"5#v"#v:V l/ t0  65"5/  ytF9$$If!vh5"5#v"#v:V l t0  65"5/  ytF9$$If!vh5"5#v"#v:V l t0  65"5ytF9$$If!vh5"5#v"#v:V l t0  65"5ytF9$$If!vh5"5#v"#v:V l t0  65"5ytF9$$If!vh5"5#v"#v:V l t0  65"5ytF9$$If!vh5"5#v"#v:V l t0  65"5ytF9$$If!vh5"5#v"#v:V l t0  65"5ytF9$$If!vh5"5#v"#v:V l t0  65"5ytF9$$If!vh5 5*#v #v*:V l/ t0  65 5*/  yt_T$$If!vh5 5*#v #v*:V l t0  65 5*/  yt_T$$If!vh5 5*#v #v*:V l t0  65 5*yt_T$$If!vh5 5*#v #v*:V l t0  65 5*yt_T$$If!vh5 5*#v #v*:V l t0  65 5*yt_T$$If!vh5 5*#v #v*:V l t0  65 5*yt_T$$If!vh5 5*#v #v*:V l t0  65 5*yt_T$$If!vh5 5#v #v:V l/ t0  65 5/  ytI@$$If!vh5 5#v #v:V l t0  65 5/  ytI@$$If!vh5 5#v #v:V l t0  65 5ytI@$$If!vh5 5#v #v:V l t0  65 5ytI@$$If!vh5 5#v #v:V l t0  65 5ytI@$$If!vh5 5#v #v:V l t0  65 5ytI@$$If!vh5 5#v #v:V l/ t0  65 5/  ytF9$$If!vh5 5#v #v:V l t0  65 5/  ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9$$If!vh5 5#v #v:V l/ t0  65 5/  ytIA$$If!vh5 5#v #v:V l t0  65 5/  ytIA$$If!vh5 5#v #v:V l t0  65 5ytIA$$If!vh5 5#v #v:V l t0  65 5ytIA$$If!vh5 5#v #v:V l t0  65 5ytIA$$If!vh5 5#v #v:V l t0  65 5ytIA$$If!vh5 5#v #v:V l/ t0  65 5/  ytF9$$If!vh5 5#v #v:V l t0  65 5/  ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9$$If!vh5 5#v #v:V l t0  65 5ytF9RDd '&0  # AbQN}M]1}ۗx?wQ noQN}M]1}ۗx?PNG  IHDRsRGBQIDATx^;VJ l8rz2ى3'!;#'xE\z-խַ=_[R# @g@@B9 @@νO!@sj = @9 @@νO!@sj = @9 @@νO!@sj = @9 @@νO!@sj = @9 @@νO!@sj = @9 @@νO!@sj = @9 @@νO!@sj = @9 @@νO!@sj =.Yq./8׀~~xwj¯uLEg!`5i LY8TDn |yVziIղv2^?5 bJ ñax؝l긒KJ]J4r2;r>YLFH'a,a2k5?-)/?]1#z$`5RE]ņNUA#>-F7yazhՏ?ȍֺt]?.6,yTwXHÍG0k"Zۈzdߐ1G4j&c X@ܪe}>pբkAW#:) :Fw|) 9VzZ0hϟVl6O:ʊ6X!0tbd۱ӽ hig:σ3J@sc`_p/u߾~+C؝ATf5]MΛo[UJ}( -RAYyeT&+Uaоn1VNS SB8ig~ּA|,v$gh;'/3ѯ\'獓Rk[=&,p99sp D'C1]S ]Ea@~ibV_ZtZ ^tո5G,:-&3VBvX2*&.W؊Q@-,-@& {4Br_>pW<^GW_9ROD0?GZ-+N\z+}), (1pozl{|үD̀aE\:yg0TCZӴO94W߳ ,[ߋ>=%+|vn~ |z b\y:snVTīrU1K\iaW K{ɍbAYhM=e s+A-a1_SM {1ƞ*PIuBZ'Wb<$Ao֞]aĄ  r.;?x@P +@  @@6A0;WD@lȹ @@r& s;@\M @ w ȹ$@d@e @ sH4 &A h@M9 (@  r.;?x@P +@  @@6\v~ @9WD@lȹ @@r& s;@\M @ w ȹ$@d@e @ sH4 &A}||(4 Ya &JU%( .<\ my+a][1@pE9wEq!@ Pb rxS}FYS{ !B#G΍!e1  Ph$, .@|&oU `@h5w:I<  p=49ɧC" ' "pvj~|,Phe) ? 8Q֥hq\BxQdƀPiʪnb9I{K@ xANK`E ' d&lW%A:X5EܓA2WE)y2Z1h/ńa@."/ޕjE$7s(4Brn#)1 m7WU֡0;_>,@jΚ3ZY#VG*s␲ 4U;Cv+zPh+f@@x^nŻK9QQ)1 `@-fmjLؗ|%Ph(6lB Os*~&/(mET<>EH-r(4DKK  3Dܻ2Ww^4 Q!0IY(!5',9SUgV/t-DܣrfBH0c@ oQqU4*H9WCʫ3ߪFȡ0;,C53__| =վ)H;vĢum}ɢPh*@gLYT\>חDȹ8l#<TLDT8frY ๐3_{])nna99l"gc ?krn/qӄ ͆!eXPQqņ;ɎCAΝdA!P#୐%QqIAmwHFx#ǬrrPh)<|K6+qC`טES됲| Y6#*nZN@@@|&?@7sSvz6z͵˧ ,7 (VF[?BF u!hS!@C&$ۙ!UF(xr9V*̷򙊇&mL=r9aȁ]OȍL\Nʍ2v|{\R9<&{c~;xm;A(xQ-Lf 66Re:.! %;r';Πs s>D,3)\|e r}'}xiȹiL6 ȓk<`Q4\tzp  4f(ZΤ~eXc #zCٹ7Qm|rzӖnC+@tȹ fh17LI9<<ZZbR %@$,)xy@έf w²Y]V7-rTo6P'om:0+4yh%)BS} ^hq` aw rح%P3sȹ' $r;E\H"pCZ/T,`RNelr2Ztz'㑷ID4rjf7ZɴAo=z|Nhy8 병!8yp)f@h6Fͤ|'rN5O-?"`R$-%Uyt2)7M~@6 r^.7ZR3o4hC`rN}xK@ϣ-)؍ŸssmMM Jh$Z.+6aRn2cL99{-^ޙo/D<r>\@]w3Ԝ"Sz^7C5! rdZC -1#1ZZA-AYR.@E9$zjgDn*6@@f{bݾD8 Zn"!,͍!z_ޟ]1볳9#30MٌPK@1 #$/iW_oh7zn$;gN/X`+_x+9?OgG_.I&?2G@xY`fމZ; rGƿF1(Z.9;|g}a v H/ߪ_OQt<|o9" σo&JA^MDMX" AΣ?ϗ]u?wˎۏ?wxKh yxD%˳( {zt$NN .O٨enF8 24 3y)$ 8 gYrN9Oa/baSs\{ʅeX{|K-ǧ`8M-|l6kT k 8<{xNN@ؙ_cfǷ JP ȑgWp\`ht -S)|< 9O|[~O[c^7.5# A}O{\={M !Ka O4 0KP2[a r H'XY왰=Z8ݎr12LA@4 r~)vw.%I-ו7bE;X@@{/gMr֞PI7BW%eUWL؁lO)r|6Rj%) ]&)bpxVؾ}$ ;-)% πE`R"$HwebNhr.ȹ.% Fɓު#F r&HrTR>GB9$B-W4-_ 0A97A5bNh`@C˨xĜ!&C$T)!9KBL2ׇI93Dr01CsC`7oV05_Yb22C py v@WLr~7!Z2hJtMȹMK+MFWAwlrdA8bnȹn@$k ȇ8 @Ԗ(| @dVFD@_lr9r^@νNeHS!#ݧ# !A ʼnD#@[OZzdot*9հjd-FFȹFrC2h2nLȹ̼xA@˗Պ-s^@?XJ@ ˒' u 0H98 -hM!o&z LT@xM@`r>͈m4-_Pk@Js%L4r3'S!3 3s H\i#Ek)@=|$5w>V|n]gl@@!r泊uf9Kc@`-|-w-hV*8!EE\_}ɦwrbVZJtrȹX1G--rL*>C0ҎrK6'Ssģh  TkDN%-WϥDIK@YBjXȹp̩J05PxC9&UEm_8D-nvȹv\Ky:A\-!6<.XJf?d6cA@8\x6SsŔ加h@79a \1OR s+ +A@΃L늠iSi+ u+?Z.ܥH< ?{E-*]8 sEN+XfW) wQ6cȹ0@L9 ŐZdFjf  Ro'D &&,޾ Z>Y23 sY' 0r> WpSJrʛ %M>@}>B@\\J9H7QN#A bZNe@0A97ALe Aȹ7 Q,PƄsy1@À'!sh@) K[0& K;3Aa9'!;reڼsPr! r.3/Ƽrh9Zn1 sJ = XƛȆpE9wEq7M-t  4B@k[ .R!( F9׆C-h9%@j@B9u *Ϧu aj2?<@KlX=@f fn]F;N@["o)ꎀk*w22 r1Snz'f1 r." 80<@"OEb<${4u2Ӓ U@M9&`jBk- ASF) z~rkKf,<'AV Z *~\]~:$맛ǫLw?>ۮy 0 @?gjzlВa^OKg^qlk|Ejz!PhFf/OQt]{U ?R,}Oo,,A((^CF\nG*z=oiwk:PCvh#,@ .϶Czr?u_D*! 4I9>y<~Es{v!վ%T\!BN 8<=NwnL[+՚y2Wlgk r[G΀ =۳Ȍ?Tqp1~R*S^{N!l71GXO''X˓y4#-@( uۻowF\`Ybg4rU-@@{S[j_QFxcj m7xh'k{|qE+}o;%/TK=-C8>NZz]L6rĖ8Q9 [NxxC`yaz| ޤG!@`{ RzWѷEۚ-25c J ycr^>^LY}fI5BWcşBMև+-H&& ,1&KD&4n F syuڡ}Wa&)$hD 0N`TCT[j?GQT[!o%  ,'{vs_>jiEg"D`^ 6Gf:f"K@3" `˅a+:SV>ٞ^JMc1pɊԄkQ=Ya3:A 8Fpx"|F$`^pۼG6>FB9JK-* ` ry /GFO@ VB"8 ȹ ʒ-\, /F@,ȹ|1>>\#z=~pj.gZe R0[+!c^b9SU 盘_[T~EZ-!@ Qt|y؝}~x6 `jNi@XI@'A+Z*vxV9y.YՄ ,9!$ yB#)=#V @`GV X1%!(px9;o i;iJָM_@@ yy#lukec/! LpxMG`;U?] [hANڃv}4D|ąF@{@>²]&(|@9O|[slϽ_-(35돎h ' A<_swc)?.,q=cBJ@ގ>-Oo'  9]׿b,MBCZiGeV2h @ |˷vw8}#" L 3oߋ$+LͽJB`3 Ӥ]ka&YX Óٹ&b̠]bR#6B9hWyzpy' t T @{ȹ2m  ȹвf](_܂ <|  ,<: -wC9ךkM J֬` @s$BM6N|gP@@匩 s)A& ! c%@E9xVڹ1^% @C9(t@u eȹ2*B@5eFǴv 436F9X B$U115wA1!s)J@`sittT6  +Ss},@`-f]oL R@˩@ 竳ZXi_ @:|&W2k/ Ts =q ]igjt@rдu @Z Z0n1K9C .J;Zh@䜺 xO9_MR7"+ +s+t/ a91Sh @""PLpgT@@Η&v{E6@::r?D^C[!;4+3$ AWSpL5  SsSd @F9_ztջ.bK'@O9ό a>>>d3VnBU]yaOah\*r_ZY"t xw~vwDf$n3ίz@F91bR# ӌtp+@Lȹ̼80< h@  'Cx`@ Xb  /:n_A_=ӝF ֙r5\tј:!狰ۉy% $r=[uiWW#wḿ?[gv?tA9k*sp@St\x_mϯOa,,_CGkݟrJ^gKt֬e>_tOٯ(m:|_[ܯho$!S/[Pﱐq3^~D}^b 9_B->L̓L+AEMjc~/Xߋd)#?]73Bx⅀Xv(]dF(otm7d}tU~ __JFڱn!Jc_=2c˽_wՈ}sQd5 B@<];GkMo_Ty|pXa.O\J=}9';2Z:2:~zyO. ªaWhm8 gd;Ʋ2\Iv܅xѳE  % =kMk1M1$\ `Fz[t>^[fbn:1rjfCIU-WD$ ac9L" CrȹRpxSsJ ܣd* @r`j0 ȹ$ _i0!@5u.oF `rn 4@@5\s1@@ \L*p ro1 xPXM00iUEc LZ^m3jz~euY Lkc(i.Z3؂0˫yO6?ؚϗ݀O7{鵀Y@pN;rݝ\=>{5,׾f<>siSs81 _S-bA]Fǫ]GQj޴(J,?L̓cw~|Ya}<QAC\t+" $Xbwt9:YyEZbP}zJj5WOQ}[SO_~=Eo_i?y1zk|09,'˭\nnx\ F@ dct1Uѓqj=c}~'_nyCF $Cf9 `כwB^%mQt٦X>wWj^L;MG}\O*e35ד]@ p|t=?/c}xnW7EoK f+9h / T*u[#&OlDo~71T.Ql&7ȹ  m+母''bwdz~y~~}N}VyAjZd9C˟LB`x)*IWO^aR{|_ӗzTy`=ڬҴW=9,;嬴35L VZi.??<+ۣkQT7.>^1sndK_K 6K&,49ׂ3h  ȧCo[*S͔x54{>:EO?EZm߾'ڿ;Djyپ>7_@`xr\םTWr?QÐ &x[gJqQwKK5j_dIݣ?XēcTxK?9Dy仟Z[C@@ y-';Lm^_Tj^L;mObj~q$y SnW>2ǟE]#ɹq>;t &mU}`ҞW}#CMj0pr=,]GM$ӄ- 25x麽$:|+z$K_kX_y7?ѣKb:1|,q]xZ%y'XJp髀; q@)Tg:1_O{xm_SO~j6r<#_U޹92ҬȽwכֿeWٶa-z;ۙߩ.;k h&ܒfkggd\>ٞ8.鉽j:On BGъgYk Y }5rsI.9.ʺ,F/FGGH&8ݶZ<;FSAjrQ`W]9R|-=/(:T9L"`:98 yУg-;$JwS*ÄT. `o-^Ǜ~n|sH |^yh (p~z4lE~KC@ &9W?S|^h =FKbp^G9&5&U8 @::r &&jSs3\ / vBձ%">jvAI7mVcWnW&=u9W9Z$A`3>FD1o_Oͷ?>uӗ0:P H.@*sIi2rj\AsAu  XY^>?<Ģ8]7~$1Ze~@a=vTMӷ=`67ȹR+G#@ ;\>W_>UNuPm);rj5WOQ~,yvx>E4V讴}oxޣ$O/R=oP{ FdTy"~!m8? a_cG'&:OQ+s'{ɕ)_Gɣ\:}=-7V͞+pX ߽i{2\\JpVxEz&Vr9L"s;)V-W%E;@&w6|f&{ &b 8)qI+[K|E)<=&9דB-Vr-1O 8TO taYlRhL  & !@Mȹ`j." 8@[ȹq@@A9w_ L xN9w@q@ H mȹ35wI!D`[KGyC@@;'lH{:IeJ$, $ @  r.! @XE9_  s Y@**|t H K>@V@W3 @@o##!@XIJt '@V@W; @=} r !@ s @pO9w< $Hw@{ȹ @`%|%@C@ @+ lU_nJtW p$ $Hr~y+,(s\ٰi,;{߳&4oQt\n$ $Hg^N~^?Tl?i,B5NvN*t $gY;~leϦ-#Y AZK!#ȂVcZP YrT)AYby6GOnC ^~Jd*QO[d魤?WOQ7Dv:Tmɂ,#?ˉF(gyrkݬFLNbg, ).hGPSYxp//Ȃ)=vwڿ> ۓ۱'d, Yx~o6JB5>lK|NfBIY؞[-$齏yU_IK^?Xu1 l@yƕyr{oI_5#Y Ajg*c(ouu8qɂ.vF<,lOΓNUN  ;y9}&3/' QKOTi>tx<-Ⅺ_ņga{r~>{7?&7_kO) RC*fm2f[ɂ:ĕ-d(r%jrUYƟ٥~l#c{y}l`b+'ϛo9[YgaȔ;idu':]9υ? e$huEHg13ߴjk" FPYhl%,y߳pǦVAC@$޽sY'@**|t H K>@V@W3 @@\B r !@  UU @@9| "Gg@ȹ, @`|>:C$@%d @  r.! @XE9_  s Y@**|t H K>@V@W3 @@\B r !@  UU @@9| "Gg@ȹ,x=;}Y.;='`ƒSWWN]`plr<% M9:gQ_Oֲyf$oKe>zs}ܴ}'"<5כÉ2o(fG럽??tz߾L.ӯCNv?s8g)?DsӄbN;ȫ[{:/v\7lY:$I'i@η^s6#OX7.Xc-<[^I1o!Qwsw \=Ƃ~|;E|-_NhwWd=nQ973_d/V [4qJ yի,y* yS7Fb?_OKm䏛W7O~<>dm>1f7@f$,!ߕ~^_ޟ''%RG35~/wW5yڪxeʍ?mt%%pߩlM(^7z}ӄ  0; @f fb `a3 @ ffb `rn6CA@p* @""l !U@EȹE @0C97 s  `rn+V!@ a3 @  WB,@-f(@ȹX X$[P 3s3\ @H9 @f fb `rn6CA@p* @""l !AjDIENDB`DyK yK vhttp://en.wikipedia.org/wiki/Statistical_poweryX;H,]ą'cDyK yK http://en.wikipedia.org/wiki/Type_I_and_type_II_errorsyX;H,]ą'c#DyK yK http://osc.centerforopenscience.org/2013/11/03/Increasing-statistical-power/yX;H,]ą'c DyK yK http://www.r-bloggers.com/power-and-sample-size-analysis-z-test/yX;H,]ą'cj 666666666vvvvvvvvv666666>6666666666666666666666666666666666666666666666666hH6666666666666666666666666666666666666666666666666666666666666666662 0@P`p2( 0@P`p 0@P`p 0@P`p 0@P`p 0@P`p 0@P`p8XV~ OJPJQJ_HmH nH sH tH J`J $*Normal dCJ_HaJmH sH tH DA`D Default Paragraph FontRi@R 0 Table Normal4 l4a (k ( 0No List 4U@4 60 Hyperlink >*phj@j  q Table Grid7:V0PK![Content_Types].xmlj0Eжr(΢Iw},-j4 wP-t#bΙ{UTU^hd}㨫)*1P' ^W0)T9<l#$yi};~@(Hu* Dנz/0ǰ $ X3aZ,D0j~3߶b~i>3\`?/[G\!-Rk.sԻ..a濭?PK!֧6 _rels/.relsj0 }Q%v/C/}(h"O = C?hv=Ʌ%[xp{۵_Pѣ<1H0ORBdJE4b$q_6LR7`0̞O,En7Lib/SeеPK!kytheme/theme/themeManager.xml M @}w7c(EbˮCAǠҟ7՛K Y, e.|,H,lxɴIsQ}#Ր ֵ+!,^$j=GW)E+& 8PK!Ptheme/theme/theme1.xmlYOo6w toc'vuر-MniP@I}úama[إ4:lЯGRX^6؊>$ !)O^rC$y@/yH*񄴽)޵߻UDb`}"qۋJחX^)I`nEp)liV[]1M<OP6r=zgbIguSebORD۫qu gZo~ٺlAplxpT0+[}`jzAV2Fi@qv֬5\|ʜ̭NleXdsjcs7f W+Ն7`g ȘJj|h(KD- dXiJ؇(x$( :;˹! I_TS 1?E??ZBΪmU/?~xY'y5g&΋/ɋ>GMGeD3Vq%'#q$8K)fw9:ĵ x}rxwr:\TZaG*y8IjbRc|XŻǿI u3KGnD1NIBs RuK>V.EL+M2#'fi ~V vl{u8zH *:(W☕ ~JTe\O*tHGHY}KNP*ݾ˦TѼ9/#A7qZ$*c?qUnwN%Oi4 =3ڗP 1Pm \\9Mؓ2aD];Yt\[x]}Wr|]g- eW )6-rCSj id DЇAΜIqbJ#x꺃 6k#ASh&ʌt(Q%p%m&]caSl=X\P1Mh9MVdDAaVB[݈fJíP|8 քAV^f Hn- "d>znNJ ة>b&2vKyϼD:,AGm\nziÙ.uχYC6OMf3or$5NHT[XF64T,ќM0E)`#5XY`פ;%1U٥m;R>QD DcpU'&LE/pm%]8firS4d 7y\`JnίI R3U~7+׸#m qBiDi*L69mY&iHE=(K&N!V.KeLDĕ{D vEꦚdeNƟe(MN9ߜR6&3(a/DUz<{ˊYȳV)9Z[4^n5!J?Q3eBoCM m<.vpIYfZY_p[=al-Y}Nc͙ŋ4vfavl'SA8|*u{-ߟ0%M07%<ҍPK! ѐ'theme/theme/_rels/themeManager.xml.relsM 0wooӺ&݈Э5 6?$Q ,.aic21h:qm@RN;d`o7gK(M&$R(.1r'JЊT8V"AȻHu}|$b{P8g/]QAsم(#L[PK-![Content_Types].xmlPK-!֧6 +_rels/.relsPK-!kytheme/theme/themeManager.xmlPK-!Ptheme/theme/theme1.xmlPK-! ѐ' theme/theme/_rels/themeManager.xml.relsPK] _   g&)$,0;?BHALQVZbdg4678?BJNQWZ]djquz{" $$$%%%y&&K**D+f+++%,R,-0&47<\=3>3??@ADGHIJ!KL?L{LMNNOKSST2UU~V5YYZ`[[q\`eg59:;<=>@ACDEFGHIKLMOPRSTUVXY[\^_`abcefghiklmnoprstvwxy|\\]]_]]]]A^_V___XXXX8@(  P  3 "?B S  ? _tryi!l!!!L"V"E#O##$&$4$S$_$p&z&-'7' ((E(N())))**++b,l,o,},-!-.//!///00?0H000:4D4G4U444@5J546=68"8>8H8V9_9::;;;;; <+<4<<<<=<=F=O=Y={==T>^>>>(?1?a?j?????@$@@@AACCE#E}GGIIJJJJ-K/K:KCKKKLLMMOOPP^PQQQQQQRRkWnWWWWWX X XXX&X1X;X_XiXlXyXzXXXXXXXXXXXXNYUYhYlYnYxYYYYYYYYYZZZ!Z)Z/Z7ZHZZZZZZZv\|\C^J^_6ZHZvZvZZZ [[Q[w[x[z[\\___- 1 k h:U. b+,^ u%d0rK_ f ""&.d,BVlT@dRFNH5KDn27TT[:b w ec6KrXJtpR| ?5}d.f,h^`OJQJo(hHh^`OJQJ^Jo(hHohp^p`OJQJo(hHh@ ^@ `OJQJo(hHh^`OJQJ^Jo(hHoh^`OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohP^P`OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohp^p`OJQJo(hHh@ ^@ `OJQJo(hHh^`OJQJ^Jo(hHoh^`OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohP^P`OJQJo(hH ^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hH^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hH^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hH^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hHh^`OJPJQJ^Jo(h^`OJQJ^Jo(hHohp^p`OJQJo(hHh@ ^@ `OJQJo(hHh^`OJQJ^Jo(hHoh^`OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohP^P`OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohp^p`OJQJo(hHh@ ^@ `OJQJo(hHh^`OJQJ^Jo(hHoh^`OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohP^P`OJQJo(hH^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hH ^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hHh^`OJQJo(hHhp^p`OJQJ^Jo(hHoh@ ^@ `OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHoh^`OJQJo(hHh^`OJQJo(hHhP^P`OJQJ^Jo(hHoh ^ `OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohp^p`OJQJo(hHh@ ^@ `OJQJo(hHh^`OJQJ^Jo(hHoh^`OJQJo(hHh^`OJQJo(hHh^`OJQJ^Jo(hHohP^P`OJQJo(hH^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hH ^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hH^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hH^`OJPJQJ^Jo(^`OJQJ^Jo(hHop^p`OJQJo(hH@ ^@ `OJQJo(hH^`OJQJ^Jo(hHo^`OJQJo(hH^`OJQJo(hH^`OJQJ^Jo(hHoP^P`OJQJo(hHlT@ecR|. 27T u%K_f "f,XJtH5K?5}&.d,- 1 [:bk                   nD                 ^.l        BC2w        7p                 l|        ]                           O        }        7p                 C6/X3%s>_ABXFCVE[FMJ\rLKNQ9R_TUDWWXhZ.:_z mHm qEqGq-v yc2zs{W}s}j4`A-=)T#K]  Gi4qWNZNuIArm.b##I%j3[u{'} RcCip; aiV79dE~PCvPw BQ>Ahy\M+!})__@HZHZ*HZHZ_h@UnknownG*Ax Times New Roman5Symbol3. *Cx Arial?= *Cx Courier New;|i0Batang7.@Calibri;WingdingsA$BCambria Math"qh'+*+DUQ0DUQ0!n20i_i_ 3QHP $P62!xxJIT51Nicole M RadziwilL           Oh+'0$ px   JIT51NormalNicole M Radziwil5Microsoft Office Word@Ik@ҦB@DUQ՜.+,D՜.+,8 hp  Hewlett-Packard0i_  Title  8@ _PID_HLINKSALK Ahttp://www.r-bloggers.com/power-and-sample-size-analysis-z-test/ OQMhttp://osc.centerforopenscience.org/2013/11/03/Increasing-statistical-power/ n7http://en.wikipedia.org/wiki/Type_I_and_type_II_errors j/http://en.wikipedia.org/wiki/Statistical_power   !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}Root Entry F0"Data ~r1TableQWordDocument<SummaryInformation(DocumentSummaryInformation8CompObjy  F'Microsoft Office Word 97-2003 Document MSWordDocWord.Document.89q