Simple random sampling examples
(→Example 1:) |
|||
| (11 intermediate revisions by 2 users not shown) | |||
| Line 1: | Line 1: | ||
| + | {{Ficontent}} | ||
===Example 1:=== | ===Example 1:=== | ||
| − | + | In this section, SRS estimators are illustrated with an example. The same example population will be used throughout the entire Lecture notes to illustrates that different [[Sampling design and plot design|sampling designs]] perform differently for the same population and with the same [[sample size]]. | |
| − | In | + | The example population has <math>N = 30</math> individual elements; we may imagine 30strip plots that cover a forest area. This dataset will also be used in the further chapters for comparison among the performance of different sampling techniques. |
| − | The example population has<math>N = 30</math> individual elements; we may imagine 30strip plots that cover a forest area | + | Table 11 lists the values of the 30 units. Here, for SRS, we are only interested in the y values. The <math>x</math> values are a measure for the size (area) of the strips; this will later be used in the context of other estimators. |
| − | Table 11 | + | |
From this population we get the following parametric values: | From this population we get the following parametric values: | ||
| − | + | :<math>\mu= \frac{\sum_{i=1}^N y_i}{N} = 7.0667</math> and | |
| + | |||
| + | :<math>\sigma^2 = \frac{\sum_{i=1}^N (y_i - \mu)^2}{N} =7.1289</math> | ||
| Line 14: | Line 16: | ||
| − | + | :<math>var (\bar {y}) = \frac {N-n}{N-1} * \frac {\sigma^2}{n} =0.491645</math> | |
| + | |||
| + | '''Table 1:'''Example Population of N = 30 individual elements | ||
| − | |||
<blockquote> | <blockquote> | ||
| − | <div style="float:left"> | + | <div style = "float:left; margin-right:4em"> |
{| class="wikitable" | {| class="wikitable" | ||
|- | |- | ||
| Line 63: | Line 66: | ||
</div> | </div> | ||
| − | + | {| class = "wikitable" | |
| − | {| class="wikitable" | + | |
|- | |- | ||
!Number | !Number | ||
| Line 106: | Line 108: | ||
|'''2087.25''' | |'''2087.25''' | ||
|} | |} | ||
| − | + | ||
| − | + | ||
| − | + | ||
</blockquote> | </blockquote> | ||
| − | |||
| − | |||
| − | |||
| − | |||
| − | |||
| − | |||
| + | This value will be compared in the subsequent chapters with the error variances produced by other [[:Category:Sampling design|sampling techniques]] with the same [[sample size]] <math>n=10</math>. The square root of the error variance is the [[standard error]]; and this is the true parametric standard error which we strive to estimate from a single sample then.To recap: the parametric standard error is the standard deviation of''all possible samples'' of size <math>n=10</math>. In a concrete sampling study, we have only one single sample of size<math>n=10</math> and from this sample the standard error can only be ''estimated''. | ||
===Example 2:=== | ===Example 2:=== | ||
| − | + | Let´stake one single sample of <math>n=10</math> from the population of <math>N=30</math> given in Figure 1 and Table1. Assume that the following elements were randomly selected: | |
| − | Let´stake one single sample of <math>n=10</math> from | + | |
| − | |||
| − | |||
| − | |||
{| class="wikitable" | {| class="wikitable" | ||
|- | |- | ||
| Line 151: | Line 142: | ||
|29 ||4 | |29 ||4 | ||
|} | |} | ||
| − | + | ||
| − | + | ||
| − | + | We now take these ten selected sampling elements to produce estimates of the population parameters of interest. The estimated mean, variance in the population and error variance are, respectively: | |
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | :<math>\bar y = \hat {\mu} = 7.1 m^3</math> | |
| − | + | ||
| − | + | ||
| − | :<math> | + | :<math>s^2 = \hat \sigma^2 = 6.9889</math> |
| + | :<math>v\hat {a}r (\bar y) = 0.4659 = \frac {N-n}{N} * \frac {s^2}{n} = fpc \frac {s^2}{n}</math> | ||
| − | + | Note that all estimated values differ from the true parametric values. In practice, however, we will never come to know how much this deviation actually is because the parametric values remain unknown. | |
| + | However,we can make a probabilistic statement about the range in which we expect the true value to be; which is the confidence interval. For the estimated mean, the [[confidence interval]] is calculated from the estimated standard error and an assumption about the distribution of the sample means which is reflected in the value of the ''t''-distribution. Accepting an error probability of <math>\alpha = 5%</math> that our statement is wrong, the width of one side of the confidence interval is: | ||
| − | |||
| + | :<math>t_{\alpha,v}S_{\bar y} = 2.262*0.6826 = 1.5440</math> and then <math>P(5.5560< \mu < 8.6440)\,</math> | ||
| − | + | ||
| − | + | ||
| + | This reads: the probability that the true parametric mean is in the interval between 5.556 and 8.644 is 0.95; it may however be that the true parametric mean is smaller or larger; this is the error probability of <math>\alpha = 5%</math>. The given ''t''-value can be read from tables or calculated from functions that usually every statistical software has built in. The actual values are depending on the degrees of freedom and the chosen error probability. | ||
| + | |||
| + | One may also calculate the confidence interval for the estimated variance in the population; which we do not here. | ||
| + | |||
| + | The above sample has a size of ''n=10'' and a [[Sampling intensity vs. sample size|sampling intensity]] of ''f= 10/30*100 = 33%''. Here, sampling intensity can be calculated in terms of number of sampling units. Usually, however, the sampling intensity in forest inventories is calculated in terms of | ||
| Line 186: | Line 174: | ||
'''Estimation of the total:''' The estimator of the total is | '''Estimation of the total:''' The estimator of the total is | ||
| − | :<math>\hat\tau = N * \bar y</math>, here: <math>\hat \tau = N * \bar y = 30 * 7.1 m^3 = 213</math> | + | |
| + | :<math>\hat\tau = N * \bar y</math>, here: <math>\hat \tau = N * \bar y = 30 * 7.1 m^3 = 213</math> | ||
and the estimated variance of the estimated total is | and the estimated variance of the estimated total is | ||
| − | :<math>\hat{var}(\hat \tau) = N^2\,v{\hat a}r(\bar y) = 30^2 * 0.4659 =419.31</math> | + | :<math>\hat{var}(\hat \tau) = N^2\,v{\hat a}r(\bar y) = 30^2 * 0.4659 =419.31</math> |
Latest revision as of 13:05, 26 October 2013
[edit] Example 1:
In this section, SRS estimators are illustrated with an example. The same example population will be used throughout the entire Lecture notes to illustrates that different sampling designs perform differently for the same population and with the same sample size. The example population has \(N = 30\) individual elements; we may imagine 30strip plots that cover a forest area. This dataset will also be used in the further chapters for comparison among the performance of different sampling techniques. Table 11 lists the values of the 30 units. Here, for SRS, we are only interested in the y values. The \(x\) values are a measure for the size (area) of the strips; this will later be used in the context of other estimators. From this population we get the following parametric values:
\[\mu= \frac{\sum_{i=1}^N y_i}{N} = 7.0667\] and
\[\sigma^2 = \frac{\sum_{i=1}^N (y_i - \mu)^2}{N} =7.1289\]
If we take samples of size \(n=10\), then the parametric error variance of the estimated mean is:
\[var (\bar {y}) = \frac {N-n}{N-1} * \frac {\sigma^2}{n} =0.491645\]
Table 1:Example Population of N = 30 individual elements
Number y x 1 2 50 2 3 50 3 6 100 4 5 100 5 6 125 6 8 130 7 6 130 8 7 140 9 8 140 10 6 130 11 7 140 12 7 150 13 9 160 14 8 170 15 10 180 16 9 200
Number y x 17 12 210 18 8 210 19 14 210 20 7 200 21 12 200 22 9 180 23 8 160 24 6 140 25 7 120 26 4 90 27 5 90 28 6 100 29 4 100 30 3 80 Mean 7.0667 13950 Pop. variance 7.1289 2087.25
This value will be compared in the subsequent chapters with the error variances produced by other sampling techniques with the same sample size \(n=10\). The square root of the error variance is the standard error; and this is the true parametric standard error which we strive to estimate from a single sample then.To recap: the parametric standard error is the standard deviation ofall possible samples of size \(n=10\). In a concrete sampling study, we have only one single sample of size\(n=10\) and from this sample the standard error can only be estimated.
[edit] Example 2:
Let´stake one single sample of \(n=10\) from the population of \(N=30\) given in Figure 1 and Table1. Assume that the following elements were randomly selected:
| Number | \(y_i\) |
|---|---|
| 3 | 6 |
| 5 | 6 |
| 9 | 8 |
| 11 | 7 |
| 15 | 10 |
| 16 | 9 |
| 21 | 12 |
| 26 | 4 |
| 27 | 5 |
| 29 | 4 |
We now take these ten selected sampling elements to produce estimates of the population parameters of interest. The estimated mean, variance in the population and error variance are, respectively:
\[\bar y = \hat {\mu} = 7.1 m^3\]
\[s^2 = \hat \sigma^2 = 6.9889\]
\[v\hat {a}r (\bar y) = 0.4659 = \frac {N-n}{N} * \frac {s^2}{n} = fpc \frac {s^2}{n}\]
Note that all estimated values differ from the true parametric values. In practice, however, we will never come to know how much this deviation actually is because the parametric values remain unknown.
However,we can make a probabilistic statement about the range in which we expect the true value to be; which is the confidence interval. For the estimated mean, the confidence interval is calculated from the estimated standard error and an assumption about the distribution of the sample means which is reflected in the value of the t-distribution. Accepting an error probability of \(\alpha = 5%\) that our statement is wrong, the width of one side of the confidence interval is:
\[t_{\alpha,v}S_{\bar y} = 2.262*0.6826 = 1.5440\] and then \(P(5.5560< \mu < 8.6440)\,\)
This reads: the probability that the true parametric mean is in the interval between 5.556 and 8.644 is 0.95; it may however be that the true parametric mean is smaller or larger; this is the error probability of \(\alpha = 5%\). The given t-value can be read from tables or calculated from functions that usually every statistical software has built in. The actual values are depending on the degrees of freedom and the chosen error probability.
One may also calculate the confidence interval for the estimated variance in the population; which we do not here.
The above sample has a size of n=10 and a sampling intensity of f= 10/30*100 = 33%. Here, sampling intensity can be calculated in terms of number of sampling units. Usually, however, the sampling intensity in forest inventories is calculated in terms of
\[f= \frac {total\, area\, of\, all\, sample\, plots}{total\, area\, of\,inventory\, region}\]
Estimation of the total: The estimator of the total is
\[\hat\tau = N * \bar y\], here\[\hat \tau = N * \bar y = 30 * 7.1 m^3 = 213\]
and the estimated variance of the estimated total is
\[\hat{var}(\hat \tau) = N^2\,v{\hat a}r(\bar y) = 30^2 * 0.4659 =419.31\]
