Module 5 Estimation Assignment

Question 1: Understanding Confidence Intervals

Confidence intervals are arguably the simplest type of statistical inference

because they allow us to use data to estimate a population parameter of
interest that we do not already know. The argument for confidence intervals
and their application in this chapter is based on the interpretation of
probability and the properties of variables and constants discussed
previously. When the boundaries are numerically determined, the population
parameter is either in the confidence interval or not, so the interval after
calculation is no longer random but is a fixed set of numbers and the
population parameter is a constant. Thereby, if the parameter is in the
interval, P is 1, or 0 otherwise.

This gives rise to an important difference in the kind of probability

statements we can make. We can ‘t really says anything about the
probability of a constant; we are limited to probabilistic statements about
random variables, since the endpoint of a confidence interval (say) are
random only with respect to the sampling procedure, not in a simulation in
which we know the statistics used to construct the interval. Prior to
sampling, the endpoints are random because their values depend on random
sample means or sample proportions, etc. Post, sampling, the random
source is gone, but the process is still a probabilistic one.

This means that for procedure ‘a’ at the confidence level c on long run,
correct intervals will be produced c percent of the time and incorrect
intervals will be produced 100, c percent of the time. When c = 95 the
resulting intervals will be correct 95 percent of the time and incorrect 5
percent of the time.

Population means can be estimated using normal distribution, the central

limit theorem and the student t distribution. If the population standard
deviation is known, or the sample size is sufficiently large the sampling
distribution of the sample mean tends to be normal. If so, then the
confidence interval for the mean can be expressed as x̄ ± critical value *
standard error. The value of the standard error is then Standard deviation /
square root of n.
For smaller samples where the population standard deviation is not known,
the student t distribution must be used rather than the normal. It extends
the normal by adding more uncertainty in the form of more probability in the
tails, hence the wider intervals. The confidence interval is then based on the
sample standard deviation and a t critical value based on degrees of
freedom, df, which is 1 less than the sample size. The t distribution tends
towards the normal as n increases.

The confidence interval for a population proportion hinge on the normal

approximation to the binomial distribution. When n p and n 1 p are large, the
sampling distribution of p hat can be treated as normal, and the following
confidence interval is formed. This is a standard technique used in polling
and survey data.

A similar approach can be used to estimate the sample size needed for a
given margin of error and a given confidence level, when the normal
distribution is used. Again, the critical value z is involved in the calculation
for the margin of error, which again equals the z, score times the standard
deviation. This formula can be rearranged to calculate the number of
observations needed to have a certain level of precision. For the population
standard deviation, use the estimated value, if the true value is not known.
If instead the z value for proportions is used, and need to be a conservative
estimate, then p should equal 0.5, because this will have the biggest
variance, and by extension the greatest sample size necessary to have a
certain level of precision.

Question 2: Interpreting Poll Results and Bias

When a sample has been used to estimate a population parameter and the
results published, the number is often reported as a point estimate plus and
minus a margin of error. For instance, in the above example a poll of 385
people in Honolulu found that 78 percent of the sample would support
requiring mandatory jail sentences for those convicted of DWI, with a margin
of error of plus or minus 3 percentage points. Build a confidence interval by
adding and subtracting the margin of error from the point estimate. Starting
with 78 percent, add and subtract 3 percent to get a confidence interval from
75 percent up to 81 percent. If the confidence level is not indicated in public
surveys, it is assumed by most people that the level is 95 percent. this is,
therefore, a 95 percent confidence interval for the population parameter of
the proportion of residents of Honolulu who support mandatory jail sentences
for individuals convicted of DWI.

It is important to keep in mind that the margin of error reflects only sampling
error and is not the same as systematic bias. Bias can come from the way
questions are written, the type of sampling used, patterns in response, and
the effects of nonresponse. For example, if a question emphasizes the
devastating injuries caused by impaired drivers, then respondents’ answers
may be biased due to wording effect. In this event, the confidence interval
would not reflect the true population proportion.

The same result applies if the question is asked about overcrowding of

prisons. Lumping the question into issues about overcrowding of prisons will
give a different form of bias, if a respondent responds to the question it may
neglect the effects of prison sentences on public safety, and instead focus on
its effects on society as a whole, thus decreasing support for mandatory jail
sentence, unless respondents prefer alternate punishments or redemption.22
While the confidence interval will in all likelihood encapsulate the correct
amount of sampling variation, it will be substantively biased nonetheless.

These examples demonstrate the difference between precision and validity.

Using Confidence Intervals researchers assess precision, I. e. The extent to
which findings were affected by random sampling error; they do not assess
validity, i. e. Whether a survey yields the data it is supposed to. Having a
low margin of error does not necessarily mean a survey is valid if the
researcher did not collect the data in a valid manner. This distinction is
particularly significant in criminal justice research, as public opinion data aid
public policy through sentencing legislation and reform and affect legislative
priorities.

To support population attitudes from surveys, survey questions should be

written in non, emotional language, sound sampling techniques utilized, and
methods clearly explained. Methodological details should be clearly reported
to prevent misinterpretations by the public that will reduce statistically, and
economically, meaningful confidence bounds.

In conclusion, confidence intervals are important for making inferences about

a population and quantifying the associated uncertainty. Correct
interpretation hinges on understanding the nature of probability, sampling
variability, and distributional assumptions. Confidence intervals are highly
useful but do not undo biases introduced by incorrect survey design.
Researchers in areas such as criminal justice need to pay close attention to
statistical as well as methodological issues to generate valid results.