Test hypothesis

Introduction
In statistics, hypothesis testing, also known as significance testing, involves

the verification of assumptions about a population parameter by analysts. The

chosen approach depends on the type of data and the study's purpose. Using

sample data, hypothesis testing assesses the credibility of a theory, whether the

data stems from a specific activity or a broader population. Throughout the

explanations that follow, the term "population" encompasses both scenarios. In

our daily lives, we often make decisions based on limited knowledge, such as

pondering whether dedicating more time to studying vocabulary will enhance

biology grades or considering becoming a chemistry major to boost chances of

admission to medical school. This section explores how hypothesis testing can

aid in making such judgments. Hypothesis testing, a form of statistical

reasoning, involves posing a question, gathering information, and evaluating

what the data suggests about the optimal course of action. Formal hypothesis

tests always present claims about the population. This chapter focuses on

hypothesis tests related to the means or average values of a variable across an

entire population. For example, one might investigate whether first-year college

students indeed spend 20 hours a week studying or compare the average study

time of freshmen with a GPA of 3.0 or above to those without. Subsequent

chapters will delve into variations in the average study time among freshmen,

sophomores, juniors, and seniors.

In hypothesis testing, analysts assess a statistical sample to substantiate

the credibility of the null hypothesis. Statistical analysts measure and scrutinize

a random sample from the target population to test a hypothesis. Each analyst

tests both the null hypothesis, which commonly posits that the population

mean return equals zero, and the alternative hypothesis, which could assert

disparities in population parameters. Essentially, the alternative hypothesis,

such as the population mean return not being equal to zero, stands in contrast

to the null hypothesis. Consequently, these hypotheses cannot simultaneously

be true, but one of them will invariably hold.

Steps of Hypothesis Testing

To scrutinize each hypothesis, a four-step process is employed:

1. The initial step involves the analyst articulating their hypotheses.

2. In the second stage, an analysis strategy is devised, outlining how the data

will be evaluated.

3. The third phase entails implementing the strategy and analyzing the sample

data.

4. The final stage involves scrutinizing the data to ascertain whether the
 evidence supports or refutes the null hypothesis.

Real-World Example of Hypothesis Testing

The alternative hypothesis contends that 50% is inaccurate, while the null

hypothesis asserts that 50% is accurate, such as when testing the assertion that

a penny has an exact 50% probability of landing on heads. The null hypothesis

can be expressed mathematically as Ho: P = 0.5, and "Ha" represents the

alternative hypothesis, denoted by the equal sign crossed out, indicating it is

not equal to 50%.

A sample of 100 coin flips is randomly taken, and the null hypothesis is

tested. If the outcome is 40 heads and 60 tails, the analyst would conclude that

a penny does not have a 50% chance of landing on heads, leading to the

rejection of the null hypothesis and acceptance of the alternative hypothesis.

Conversely, if there are 48 heads and 52 tails, it is conceivable that the

coin could be fair and still yield such results. In instances where the null

hypothesis is "accepted," indicating no significant difference between expected

(50 heads and 50 tails) and observed results (48 heads and 52 tails), the analyst

attributes the difference to chance alone.

Developing Null and Alternative Hypotheses

There are always two hypotheses in a statistical hypothesis test. The null
hypothesis, or H0, is the hypothesis that has to be tested. According to the null

hypothesis, there is no distinction between the sample mean and the

population mean that is being conjectured. It is the hypothesis of the status

quo. To test the hypothesis that first-year college students study for 20 hours a

week, for instance, we might formulate our null hypothesis as follows: H0: µ =

20. We compare the alternative hypothesis, represented by the symbol Ha,

with the null hypothesis. The theory that you personally believe in is frequently

the alternative hypothesis! It encompasses the results that are not addressed

by the null hypothesis. Our alternative hypothesis in this case would state that

first- year students do not study for 20 hours a week : Ha : µ 6= 20

For a medicine being manufactured, where each pill is intended to contain 14

milligrams of the active ingredient, the null and alternative hypotheses are as

follows:

Null Hypothesis (H0): µ = 14

Alternative Hypothesis (Ha): µ ≠ 14

The null hypothesis asserts that the population mean is equal to 14

milligrams, while the alternative hypothesis suggests that the population

mean differs from 14 milligrams.

Determining Whether to Reject the Null Hypothesis:

In the context of one and two-tailed hypothesis tests, rejecting the null

hypothesis is the means by which we lend support to the alternative

hypothesis. Rejecting the null hypothesis entails discovering a significant

difference between the sample mean and the proposed (null) mean, casting

doubt on the idea that the authentic population mean is 20.

To reject the null hypothesis, there must be a substantial disparity

between the sample mean and the hypothesized mean. If the difference is
 negligible, the null hypothesis is not rejected. It is crucial to predefine the

magnitude of difference required for each hypothesis test to make informed

decisions regarding null hypothesis rejection.

In summary, the null hypothesis cannot be rejected unless there is a

significant difference to substantiate rejection, leaving no alternative options

in this decision-making process.

A statistician needs to establish the magnitude of a mean difference required

to reject the null hypothesis during hypothesis testing. Initially, statisticians set

the alpha (α) level, or significance level, for a hypothesis test. This alpha level

signifies how improbable a sample mean must be to be deemed "significantly

different" from the hypothesized mean, similar to the significance level used for

generating confidence intervals. The most commonly utilized significance values

are 0.05 and 0.01. If the probability of encountering the sample mean is below

5%, we consider our sample mean to be significantly different from the

level of 0.01, we regard our sample mean as significantly different from the

hypothesized mean if the likelihood of observing that sample mean is less than

1%.

Two-tailed Hypothesis Tests

previously mentioned scenarios, we are dealing with two-tailed hypothesis

tests. In these tests, we assert either that an average of 20 hours per week is

spent studying or that it is not, similarly for an average of 3.2 hours per week

spent on computers. We do not specify whether we believe the true mean is

higher or lower than the hypothesized mean; instead, we simply posit that

there is a discrepancy.

In a two-tailed test, if the sample mean falls into either tail of the

distribution, the null hypothesis is rejected. Consequently, the alpha level, let's

say 0.05, is distributed across both tails. The accompanying curve illustrates the

critical regions for a two-tailed test, with each tail having a probability of 0.025.

The critical values, which are the z-scores marking the beginning of these

regions, help identify if the sample mean falls within the critical or "rejection"

regions. If it does, suggesting a significant difference, we reject the null

hypothesis. Conversely, if the sample mean falls between these critical regions,

indicating no significant difference, we fail to reject the null hypothesis.

In cases where the direction of findings can be predicted or when interest

is focused in a specific direction, a single-tail hypothesis test is appropriate. An

example of a single-tail hypothesis test could be in the evaluation of whether to

adopt a new textbook. The decision to adopt the textbook would be contingent

on it leading to higher student achievement than the previous one.

In a single-tail hypothesis test, the alternate hypothesis is expressed

differently, utilizing symbols such as greater than or less than. For instance, if we

assert that the mean SAT score of graduating seniors is HIGHER than 1,110, the

alternative and null hypotheses would be formulated as follows:

\[ H_0 : \mu \leq 1100 \]

\[ H_a : \mu > 1100 \]

Here, the null hypothesis posits that the mean SAT scores are less than or

equal to 1,100, while the alternative hypothesis suggests that the SAT scores
exceed 1,100. A single-tail hypothesis test involves a solitary critical region,

concentrating the entire critical region on one side of the distribution. When the

alternative hypothesis states that the sample mean is greater, the critical region

is situated on the right side of the distribution. Conversely, when the alternative

hypothesis implies a smaller sample mean, the critical region is located on the

left side of the distribution.

To calculate the critical regions, we must first find the critical values or the cut-

offs where the critical regions start. This will be covered in the next section.
Type I and Type II Errors

Keep in mind that there will be some sample means that are outliers – this

occurrence is expected to happen approximately 5% of the time, given that

around 95% of all sample means fall within about two standard deviations of

the mean. What happens if, during a hypothesis test, we obtain an outlier

sample mean? Even if it originates from that distribution, it won't resemble our

hypothesized mean. In such cases, we might be inclined to reject the null

hypothesis, but that decision would be incorrect.

When determining whether to reject or not reject the null hypothesis, we face

four potential scenarios:

a. Rejecting a true hypothesis.

(Option 2) or rejecting a false hypothesis (Option 4). However, if we reject a true

hypothesis (Option 1) or fail to reject a false hypothesis (Option 3), we commit

an error. In essence, neither type of error is inherently more severe than the

other. The seriousness of each type depends on the specific research context,

but the ideal scenario is to minimize both types of errors during the analysis.
Conclusion

The general approach to hypothesis testing centers on the Type I

error, which involves rejecting the null hypothesis when it might be true.

Interestingly, the level of significance, also known as the alpha level,

represents the probability of committing a Type I error. At the 0.05 level,

the decision to reject the hypothesis may be incorrect 5% of the time.

Calculating the probability of a Type II error is more intricate than

determining a Type I error, and we won't delve into that here. It's essential

to recognize each type of error in a specific hypothesis test. For instance,

consider testing whether listening to rock music enhances memory recall

for 30 random objects, assuming it does not. A Type I error would involve

incorrectly concluding that listening to rock music improved memory. This

type of error only occurs when the null hypothesis is false. Conversely,

assuming that listening to rock music does enhance memory, if one

incorrectly concludes that it doesn't, it constitutes a Type II error—failing to

detect a significant difference when one actually exists.

Furthermore, it's crucial to acknowledge that the likelihood of a Type

I error is within our control. The alpha level is often established based on
the severity of the consequences associated with making a Type I error. If

the repercussions are not overly serious, one might set an alpha level at

0.10 or 0.20, indicating a willingness to make a decision where falsely

rejecting the null hypothesis is acceptable 10 to 20% of the time. However,

in fields like medical research with potential harm to patients, a very low

alpha level, such as 0.001, might be set to minimize the risk of Type I errors.