CHI-SQUARE AS A TEST OF ‘GOODNESS OF FIT’

The difference between observed and expected frequencies is expressed in terms of a statistic named
chi-square(𝜒2), introduced by Karl Pearson in 1900.
As a test of ‘goodness of fit’, Karl Pearson tried to make use of the χ² distribution for devising a test for
determining how well the experimentally obtained results fit in the results expected theoretically on some
hypothesis.
CHI-SQUARE AS A TEST OF ‘GOODNESS OF FIT’ implies using chi-square distribution for devising a
test for determining how well the experimentally obtained results fit in the results expected theoretically on
some hypothesis.
The Goodness-of-Fit test is a statistical method used to check whether the data you have observed
(collected from an experiment or survey) matches what you expected to happen, based on a theory or
assumption about the population. It is mostly applied when the data is grouped into categories (qualitative
data), and you want to see if the number of observations in each category fits well with the numbers you
predicted under some hypothesis.
In this test, the Chi-Square (𝝌²) statistic is used to measure the difference between the observed
frequencies and the expected frequencies.

• The observed frequency (fₒ) refers to the actual number of times an outcome or category appears in
your sample. For example, if you roll a die 60 times and get "3" on 8 occasions, then 8 is the
observed frequency for that category.

• The expected frequency (fₑ) is the number of times you would expect that outcome to occur, if your
assumption (called the null hypothesis) is correct. For example, for a fair die, each number (1 to 6)
should occur equally often, so the expected frequency for each number in 60 rolls would be 10. The
expected frequency can also be thought of as the average number of times you would observe each
outcome if you repeated the experiment an infinite number of times under the same conditions,
assuming the null hypothesis is true.
Depending on the nature of the expected distribution, the Chi-Square test can be applied under
various hypotheses — most commonly: hypothesis of chance, hypothesis of equal probability,
and hypothesis of normal distribution.
1. Hypothesis of Chance
The first use of the Chi-Square test is under the hypothesis of chance, which checks whether observed
outcomes differ significantly from what is expected by random probability. For example, in tossing a fair
coin 100 times, we expect 50 heads and 50 tails. If we observe 40 heads and 60 tails, the Chi-Square test
helps determine if this difference is due to chance or something more. It’s also used in dice rolls or
guessing on multiple-choice tests, where expected outcomes are based on known probabilities. This
hypothesis is useful for testing if results align with random chance.
2. Hypothesis of Equal Probability
The hypothesis of equal probability assumes all outcomes have the same chance of occurring. The Chi-
Square test checks if observed data fits this assumption. For example, in a survey with 60 participants
choosing between "Yes," "No," and "Indifferent," we expect 20 responses in each. If the actual counts
differ, like 22 "Yes," 18 "No," and 20 "Indifferent," the test shows whether this variation is significant. This
method is useful in surveys, polls, and studies where equal response likelihood is assumed.
3. Hypothesis of Normal Distribution
The third and more complex use of the Chi-Square test involves comparing observed data to a normal
distribution. Expected frequencies are based on the bell-shaped normal curve rather than equal or
chance-based values. For example, in a personality test given to 200 people, categories like "Very good"
to "Very poor" should follow known normal curve percentages (e.g., 3.5%, 23.84%, 45.14%, etc.). The
Chi-Square test checks if the observed scores align with these expected values, making it useful in fields
like psychology and education.
The difference between the observed and expected values is measured by the
Chi-Square formula:
This formula calculates how much the observed data deviates from the expected data, and whether the
difference is small (likely due to chance) or large (likely not due to chance).

• = observed frequency (what actually happened in the experiment)

• = expected frequency (what you assumed would happen if the hypothesis is true)

• ∑ means you perform this calculation for each category, and then add them all together.
If the calculated 𝝌² value is small, it suggests that the difference between observed and expected
frequencies is likely due to random variation, so your hypothesis may still be correct.
If the 𝝌² value is large, it suggests the observed data does not fit the expected pattern, and you may have
to reject the hypothesis.
These points explain key concepts about the chi-square (𝝌²) statistic, which is commonly used in
statistical tests to measure how well observed data fits an expected distribution. Here's what each point
covers:
1. 𝝌² Can Never Be Negative: The chi-square value is always zero or positive because it’s based on
squaring differences, making negative values impossible.
2. 𝝌² is Zero Only When Observed = Expected: 𝝌² is zero when the observed values perfectly match
the expected ones, indicating no difference.
3. Larger Differences Mean Larger 𝝌²: A bigger difference between observed and expected values
leads to a higher chi-square statistic, suggesting stronger evidence against the null hypothesis.
4. Relative Difference is More Important: It's not just the size of the difference, but how big that
difference is in relation to the expected value. This adjustment makes the statistic more meaningful.
5. Degrees of Freedom Affect 𝝌²: The degrees of freedom (df) are based on the number of categories
being compared, and more categories mean more opportunities for differences to occur.

• Chi-Square Distribution
The chi-square distribution is a mathematically defined curve that represents the possible values of the
chi-square statistic (𝝌²) under the assumption that the null hypothesis is true.
Whenever you calculate a chi-square value from your observed and expected frequencies, you compare
it to this distribution to decide whether the difference is "too big" to have happened by random chance.

Key points about the chi-square distribution:

• It is not symmetrical — the curve is skewed to the right.

• It is always positive (because the formula uses squared differences).
• Its shape depends on the degrees of freedom (df) — the more degrees of freedom, the more the
distribution shifts to the right and becomes more spread out.
• It helps to calculate the probability of getting a particular chi-square value (or larger) if the null
hypothesis is true.
Chi-Square Table
The chi-square table lists specific cutoff values (also called critical values) from the chi-square
distribution for different situations.
It helps you decide whether your calculated chi-square value is "significant" — meaning whether the
observed difference is unlikely to have occurred by random chance.
The table is organized by:

• Degrees of freedom (df):

This depends on the number of categories in your problem. Usually calculated as: df=
number of categories−1

• Significance levels (α):

This shows how strict your test is — common values are 0.05 (5%) or 0.01 (1%).
Interpretation of the Outcome of a Chi-Square Test:

• Significance of Chi-Square (𝜒²): When the chi-square value is significant (larger than the critical
value), it tells us that there is a statistically significant difference between observed and expected
frequencies, leading us to reject the null hypothesis (H₀).

• Limitations: Even though we reject H₀, the chi-square test doesn't specify which particular
preferences are involved. There could be multiple reasons behind the rejection of H₀, such as
preferences for multiple brands (e.g., B and C) over others.

• Directional vs. Non-Directional Tests: A directional test is used when there are only two possible
outcomes for how the null hypothesis (H₀) can be wrong, such as when comparing two categories
(like two brands). In such cases, the chi-square test (𝜒²) can be treated like a z-test (because 𝜒² = z²
when there are two categories, df = 1). This allows you to calculate a z-value and compare it with
critical z-values to see if the result is significant in one specific direction.
Assumptions for Using Chi-Square Distribution:
The chi-square test assumes certain conditions, and if these are not met, the results may not follow the
chi-square distribution accurately:
1. Random Sampling: The sample should be randomly drawn from the population you wish to make
inferences about. In practice, this assumption is rarely perfectly satisfied.
2. Independence of Observations: Observations should be independent. For example, in the soft drink
study, if each subject had multiple trials, their responses would not be independent. If there are more
observations than subjects, independence may be compromised.
3. Normal Distribution of Observed Frequencies: It’s assumed that, over repeated experiments, the
observed frequencies will follow a normal distribution around the expected frequencies. This tends to
be true when random sampling is used.
BUILDING A CHI-SQUARE (Χ²)
1. Establishing the Null Hypothesis:
It may be stated that there exists no actual difference between the observed frequencies (derived from
experimental results) and expected frequencies (derived on the basis of some hypothesis of equal
probability or chance factor or hypothesis of normal distribution).
2. Computation of the χ² Statistic:
The next step involves calculating the χ² statistic using the formula:

▪ = Observed frequency (the actual counts from the data).

▪ = Expected frequency (the theoretical values you expect based on the

null hypothesis).
For each category, you compute the difference between the observed and expected frequencies, square
it, and divide by the expected frequency. Then sum these values for all categories to get the χ² value.
3. Determining Degrees of Freedom (df):
Degrees of freedom are calculated based on the structure of the contingency table (if using a contingency
table). The formula for degrees of freedom is:
df= (r−1)(c−1)
where r is the number of rows and c is the number of columns in the table.
4. Determine the Critical Value from the χ² Distribution Table:
Using the degrees of freedom (df) and the desired significance level (usually 0.05 or 0.01), you refer to a
χ² distribution table to find the critical value that corresponds to your test.
5. Compare the Computed χ² Value to the Critical Value:
Finally, we compare the computed χ² value with the critical value obtained from the table:
▪ If the computed χ² is greater than or equal to the critical value, the result is considered statistically
significant, and you reject the null hypothesis.
▪ If the computed χ² is less than the critical value, the result is not significant, and you fail to reject the
null hypothesis, meaning the observed differences could simply be due to chance.
This process is used to test the goodness of fit (whether the data follows a specific distribution) or test the
independence of categorical variables (such as whether two variables are related).
ASSUMPTIONS OF CHI-SQUARE TEST
1) Frequencies or Proportions:
The Chi-square test is used when data is expressed in the form of frequencies—how often something
occurs. Even if the data is originally in percentages or proportions, it must be converted into actual counts
(frequencies) for analysis.
2) Categorical or Discrete Data:
This test is suitable for categorical or discrete data, such as gender, blood type, or preference categories.
If the data is continuous (like height or weight), it must be grouped into intervals or categories before
applying the Chi-square test.
3) No Normality Assumption:
The Chi-square test does not require the data to follow a normal distribution, unlike parametric tests such
as the z-test or t-test. This makes it a non-parametric test, useful for data that doesn’t meet normality
conditions.
4) Independence of Observations:
Each data point must come from a different individual or source, meaning the outcome of one observation
should not influence another. This assumption ensures the results are not biased due to related or
repeated measures.
5) Matching Totals:
The total sum of expected frequencies should always be equal to the total sum of observed frequencies.
This ensures that the comparison between observed and expected values is valid and statistically
meaningful.
USES OF CHI-SQUARE TEST
1) Goodness of Fit:
The Chi-square test for goodness of fit evaluates whether the distribution of observed frequencies
matches the expected frequencies based on a specific theoretical model or hypothesis. It helps to
determine how well a set of observed data fits a particular distribution.
2) Test of Independence:
This use of the Chi-square test helps assess whether there is a significant association between two
categorical variables in a population. It tells us if the variables are related or if they occur independently of
each other.

LIMITATIONS OF CHI-SQUARE TEST

1. Small Frequencies in 2×2 Tables: When expected frequencies are below five in 2×2 tables (df = 1),
Yates’ correction is needed to adjust the chi-square value.
2. Multiple Degrees of Freedom and Small Cells: With more than one degree of freedom, small cell
frequencies may distort results, requiring categories to be combined, which can reduce accuracy and
experimental precision.

ANALYSIS OF VARIANCE (ANOVA)

Analysis of variance (ANOVA) is a hypothesis-testing procedure that is used to evaluate mean differences
between two or more treatments (or populations). As with all inferential procedures, ANOV A uses sample
data as the basis for drawing general conclusions about populations.
The method was first introduced by Sir Ronald Fisher in the 1920s, providing a powerful way to solve
problems involving comparisons of multiple groups. ANOVA tests whether the variation between group
means is larger than would be expected by chance, based on the variation within the groups.
There are two possible interpretations of the observed differences in sample means:
1. Null Hypothesis Interpretation (H₀):
This interpretation suggests that there are no real differences between the populations (or
treatments). Any observed differences in the sample means are due to random, unsystematic
factors or sampling error. Essentially, this means that the variation observed between the group
means can be attributed to chance, not to any actual difference between the groups.
If the ANOVA results fail to reject the null hypothesis (i.e., the p-value is greater than the significance
level), it suggests that the observed differences are likely due to random sampling errors, and the groups
are not significantly different from each other.
2. Alternative Hypothesis Interpretation (H₁):
This interpretation suggests that the populations (or treatments) really do have different means,
and the differences between the sample means are systematic rather than random. In other words,
the differences are not due to chance, but rather reflect true variations in the populations or
treatments being compared.
If the ANOVA results reject the null hypothesis (i.e., the p-value is less than the significance level), it
indicates that there are statistically significant differences between the group means, meaning that at
least one group is different from the others.
The Purpose of ANOVA
Analysis of Variance (ANOVA) is a statistical method used to compare the means of three or more
groups. It helps determine whether the differences observed among sample means are due to a real
effect or simply due to chance. Instead of running multiple t-tests, which increases the risk of Type I error,
ANOVA allows a single test to assess all group differences together.
Characteristics

• The Null and Alternative Hypotheses- The null hypothesis (H₀) in ANOVA states that all group
population means are equal (e.g., 𝜇A = 𝜇B = 𝜇C). This means the treatments have no effect. The
alternative hypothesis (H₁) suggests that at least one group mean is different. ANOVA is called an
omnibus test because it tests for any overall difference, without specifying which group differs.
• Variance Instead of Means-Although ANOVA is used to compare means, it works by analyzing
variances. This may seem odd, but variance provides a way to estimate how much the sample
means differs due to treatment effects versus random chance. So, the method focuses on comparing
two sources of variation: within groups and between groups.
• Within-Groups Variation (Error Variance)- Within-groups variation reflects the variability among
individuals within the same group. This type of variation is due to chance factors—such as
individual differences or measurement error—and is not influenced by the treatment. This estimate
remains consistent whether or not the null hypothesis is true and is used as a baseline for
comparison.
• Between-Groups Variation (Treatment + Error)- Between-groups variation reflects differences
among group means. This variation can come from two sources: inherent random differences and
actual treatment effects. If the null hypothesis is true, the differences among means are due only to
chance. If H₀ is false, the variation also includes real differences caused by the treatment.
The Logic Behind ANOVA
The core logic of ANOVA is this: if the between-groups variation is significantly larger than the within-
groups variation, we conclude that at least one group mean differs due to a real effect. This is tested
using the F-ratio, which is the ratio of between-groups variance to within-groups variance. A larger F
indicates greater evidence against the null hypothesis.

• THE COMPARISON BETWEEN THE T-TEST AND F-TEST (ANOVA)

Conceptual Relationship:
The t-test and F-test are closely related, particularly when comparing two groups. In fact, for the special
case of two groups, ANOVA (F-test) and the t-test lead to exactly the same conclusions. Both tests are
used to assess whether the means of different groups are significantly different from each other. The t-
test is typically used when comparing two independent sample means, whereas the F-test (ANOVA) is
used when comparing three or more groups.
Formula and Calculation:
The formula for the t-test involves calculating the difference between sample means, divided by the
estimated standard error of the difference between the means. This helps in testing whether the
difference observed in sample means is statistically significant.

• t = (difference between sample means) / (estimated standard error of the difference)

On the other hand, the formula for the F-test compares variance estimates. Specifically, the numerator
of the F formula represents the variance estimate based on the differences between sample means, while
the denominator represents the variance within the samples (i.e., the variability of individual scores
within each group).

• F = (variance between sample means) / (variance within samples)

Concept of Variance vs. Standard Deviation:
The t-test uses the standard deviation (a measure of spread or dispersion) of the sample data in its
calculations, which is a direct measure of the average distance between data points and the mean. In
contrast, the F-test uses variance, which is simply the square of the standard deviation. By squaring the
standard deviation, ANOVA measures the variability of the data in relation to the group means, which
makes the F-test suitable for comparing multiple group variances.
Scope of Application:
While the t-test is generally limited to comparing only two groups, the F-test in ANOVA extends this
comparison to three or more groups. This flexibility allows ANOVA to test for differences among multiple
groups simultaneously, controlling the risk of Type I errors (false positives) that would increase if multiple
t-tests were conducted. If ANOVA detects significant differences, follow-up tests (post-hoc tests) can
identify which specific group pairs differ.
Interpretation of Results:

• In the t-test, the outcome is a t-statistic and the associated p-value, which tells us whether the
difference between the two groups' means is statistically significant.

• In the F-test (ANOVA), the outcome is an F-statistic and its p-value. If the p-value is small (typically
less than 0.05), it indicates that there are significant differences among the group means. However,
ANOVA only tells us that at least one group is different from the others, not which groups are different,
necessitating further analysis (post-hoc tests).
In summary, while both tests aim to assess mean differences, the t-test is more suited for comparing two
groups, while the F-test (ANOVA) is ideal for comparing multiple groups and controlling error rates.
ASSUMPTIONS OF ANOVA
1. Normality of Populations:
One of the primary assumptions of ANOVA is that the populations being compared are normally
distributed. This means that the data from each group should follow a bell-shaped curve, with most values
clustered around the mean. However, moderate deviations from normality generally do not significantly
affect the results of ANOVA, especially when the sample size is large. The larger the sample size, the
more robust the ANOVA test becomes to slight non-normality in the data.
2. Homogeneity of Variance (Equal Variances):
ANOVA assumes that the variance (or spread) of the scores in each group is equal, known as
homogeneity of variance. This means that the amount of variation within each group should be similar.
If the variances are very different between groups, the results of the test can become unreliable,
especially when sample sizes differ. However, when the sample sizes are equal across groups, slight
violations of this assumption are less problematic.
3. Independence of Samples:
Independence refers to the idea that the data points within each group and between groups should be
independent of one another. In other words, the data collected from one group should not influence the
data in any other group. Violating this assumption—such as in cases where the same subjects are
measured multiple times (repeated measures)—invalidates the results of ANOVA. For such cases, a
repeated measures ANOVA is used instead, which accounts for the correlated nature of the data.
4. Random Sampling with Replacement:
The assumption of random sampling with replacement ensures that each sample is drawn randomly
from the population, allowing for generalization of results. In practice, most researchers use random
assignment, where participants are randomly assigned to different groups rather than sampled from the
population. While random sampling is ideal, random assignment is often sufficient for conducting valid
ANOVA tests. Random sampling ensures the ability to extend conclusions from the sample to the larger
population.

• POST HOC COMPARISONS

Post hoc comparisons are statistical tests conducted after obtaining a significant result in an analysis of
variance (ANOVA). When the null hypothesis (which states that there are no differences between the
population means) is rejected, it indicates that there are significant differences somewhere between the
groups, but ANOVA alone doesn’t tell us exactly where those differences lie. This is where post hoc
comparisons come into play—they allow us to determine which specific group means are significantly
different from one another.
Why Are they Necessary?
In an ANOVA, we typically test if there are any differences among the group means as a whole. However,
if the overall F ratio is significant, the next step is to investigate which groups differ from each other. Since
ANOVA does not provide specific details about the differences between individual means, post hoc tests
examine all possible pairwise comparisons of the group mean to identify where the significant differences
exist.
Types of Post Hoc Tests

• Tukey’s Honestly Significant Difference (HSD): A popular test that compares all possible pairs of
means. It is less conservative than some other tests, meaning it has higher power to detect true
differences.
• Scheffé Test: A more conservative test, reducing the risk of Type I errors but lowering statistical
power.
• Duncan’s Multiple Range Test and Newman-Keuls Test: These are less conservative and more
likely to identify differences but increase the risk of Type I errors.
Dealing with Multiple Comparisons
Post hoc tests are designed to prevent Type I errors, which occur when a true null hypothesis is
incorrectly rejected (false positives). When making many comparisons (for example, testing all pairs of
means), the chance of finding a statistically significant result by chance increases. To control for this, post
hoc tests adjust the significance level or use methods like the Bonferroni correction, which divides the
overall alpha level (e.g., 0.05) by the number of comparisons being made. However, this can make it
harder to detect true effects, as it reduces the power of the test.
Concerns About Post Hoc Tests
While post hoc tests are useful, some statisticians caution against their overuse because they can lead to
"fishing expeditions". This refers to the practice of exploring data after the fact without prior hypotheses,
which increases the risk of finding spurious (false) results. Statisticians recommend that researchers only
use post hoc tests when they have clear, well-defined hypotheses or when the study design necessitates
them.
Additionally, testing all possible pairs of means often results in overly conservative conclusions that may
fail to detect true differences. Therefore, some statisticians suggest focusing on specific comparisons that
are of theoretical or practical interest rather than conducting broad post hoc analyses.
In conclusion Post hoc comparisons are an essential follow-up to a significant ANOVA result. They help
identify where the differences between groups lie, but they must be used cautiously to avoid inflating the
risk of Type I errors. By adjusting significance thresholds and using conservative methods, researchers
can ensure that their conclusions are both meaningful and reliable.