Understanding Chi-Square ($\chi^2$) Test Applicability

The Chi-Square ($\chi^2$) test is a statistical tool used primarily to examine the relationship between categorical variables. It assesses how well the observed frequencies in different categories fit the expected frequencies. However, like any statistical test, the $\chi^2$ test has specific conditions that must be met for its results to be reliable. Violating these conditions can lead to inaccurate conclusions.

Conditions for Applying the Chi-Square Test

The validity of the $\chi^2$ test, particularly the Pearson's chi-square test, relies heavily on the distribution of expected frequencies within the data categories. The key conditions include:

• The data must be in the form of frequencies or counts for categorical variables.
• The observations must be random and independent.
• A crucial rule relates to the expected frequencies (E) in each cell of the contingency table. Generally, the $\chi^2$ test is considered reliable when:
  • No expected frequency is less than 1.
  • No more than 20% of the cells have an expected frequency less than 5.

When these expected frequency conditions are not met, the $\chi^2$ distribution approximation becomes inaccurate, and the test results may be misleading.

Analysis of Provided Options

Let's evaluate each option based on the standard conditions for the $\chi^2$ test:

• Option 1: If there are only two cells, expected frequency in each cell should be 5 or more.

  This statement is partially correct, especially for 2x2 contingency tables where Fisher's Exact Test is often preferred if expected frequencies are low. However, the general rule for larger tables is the 20% guideline mentioned above. If a 2x2 table has expected frequencies below 5, corrections like Yates' correction are sometimes used, or alternative tests are recommended. But this is a specific case, not the general condition where the test *should not* be applied in all circumstances.

• Option 2: Samples of at least 50 observations must be drawn randomly from the population.

  While random sampling is essential for statistical inference, there isn't a fixed minimum sample size like 50 required for the $\chi^2$ test itself. The primary requirement is met regarding expected frequencies. A very small sample size might lead to low expected frequencies, but the condition itself isn't a strict minimum sample size requirement.

• Option 3: If number of cells are more than two and more than 20% of the cells have expected frequency less than 5.

  This option accurately describes a situation where the $\chi^2$ test assumption regarding expected frequencies is violated. When more than 20% of the cells have an expected frequency below 5, the accuracy of the $\chi^2$ statistic as an approximation of the true distribution is compromised. Therefore, the $\chi^2$ test should not be applied in this scenario.

• Option 4: Data should be expressed in original units rather than ratio or percentage.

  The $\chi^2$ test is specifically designed for analyzing frequencies or counts of categorical data. Ratios or percentages can be used if they represent proportions within categories, but the core data analysed are counts. This statement suggests a preference for raw data units, which is more relevant for tests dealing with continuous variables, not categorical frequency data typical for $\chi^2$ tests.

Conclusion on Chi-Square Test Applicability

The Chi-Square ($\chi^2$) test is a powerful tool, but its reliability hinges on meeting specific data conditions. The most critical condition relates to the expected frequencies in each category. When a significant proportion of cells (more than 20%) have low expected frequencies (less than 5), the mathematical assumptions underlying the test are violated. This directly impacts the accuracy of the test results, making it inappropriate to use the $\chi^2$ test under such circumstances.

Therefore, the condition where the $\chi^2$ test should not be applied is when more than 20% of the cells have expected frequencies less than 5, provided there are more than two cells.