Internal consistency refers to how well a survey, questionnaire, or test actually measures what you want it to measure.
The higher the internal consistency, the more confident you can be that your survey or test is reliable.
One popular way to measure internal consistency is to use split-half reliability, which is a technique that involves the following steps:
1. Split a test into two halves. For example, one half may be composed of even-numbered questions while the other half is composed of odd-numbered questions.
2. Administer each half to the same individual.
3. Repeat for a large group of individuals.
4. Find the correlation between the scores for both halves.
The higher the correlation between the two halves, the higher the internal consistency of the test or survey. Ideally you would like the correlation between the halves to be high because this indicates that all parts of the test are contributing equally to what is being measured.
When to Use Split-Half Reliability
The split-half reliability method is an easy method to carry out if you want to measure internal consistency, but it should only be used if the following two conditions are present:
1. The test has a large number of questions. Split-half reliability works best for tests that have a large number of questions (e.g. 100 questions) because the number we calculate for the correlation will be more reliable.
2. All of the questions on the test or survey measure the same construct or knowledge area. If a particular test measures several different constructs like leadership skills, communications skills, programming skills, and other professional skills, then a split-half reliability would not be appropriate to use since many of the responses are not expected to be correlated anyway.
Example of Split-Half Reliability
Suppose researchers would like to measure the internal consistency of a particular test that has 100 questions all related to introverted personality traits. They carry out the following steps to measure the split-half reliability of the test.
Step 1. Split the test in half based on odd-numbered and even-numbered questions.
Step 2. Administer each half of the test to the same individual.
Step 3. Repeat for 50 individuals.
Step 4. Find the correlation between the scores for both halves.
If the researchers find that the correlation is fairly high, they can be assured that all parts of the test are contributing equally to measuring the introverted personality traits they’re interested in.
Conversely, if the correlation is low then that could be an indication that the lowly correlated questions need to be either re-written or removed altogether to improved the internal consistency and the overall reliability of the test.
Additional Resources
The following tutorials provide additional information about reliability in statistics:
A Quick Introduction to Reliability Analysis
What is Test-Retest Reliability?
What is Inter-rater Reliability?
What is Parallel Forms Reliability?
What is Standard Error of Measurement?
the first line of your post is misleading, “Internal consistency refers to how well a survey, questionnaire, or test actually measures what you want it to measure.” This is false. Internal consistency refers to how highly correlated the items of a test are. Just because test items are internally consistent, does NOT mean they measure what you want them to. They measure something, but we need to use construct validation principles to understand what exactly is measured. Reliability does NOT equal validity.
Hello,
I think that you have done a good job indicating how to physically split a test into two halves of obtain a correlation and thus a measure of reliability. However, you have not discussed the aspect that you will have a correlation based on half the number of items. Thus, a ten item test will be split into two halves consisting of 5 items each. The correlation will then give a measure of reliability for a five-item test. Typically, researchers would apply the Spearman-Brown prophecy formula (Gutmman and Rulon have proposed other formulas if the variability varies across the two halves) to determine what would be expected from a ten-item test (doubling the length of the test).
To make your story more complete, I think that you ought to add that additional information.
Can I do split half method for Metric variables ( Volume Data in liters 30 measurements each, and pressure data , 30 measurements) ?
Which Correlation should I use for metrical Data for a split half reliability test ?
Good .Kindly elaborate step
4. on determining correlation after splitting the question into 1 and 2 half
Hi Kirema…### Step 4: Determining Correlation After Splitting the Questions into Two Halves
Split-half reliability assesses the internal consistency of a test by dividing its items into two subsets and comparing the results. After splitting the test into two halves, the next step is to determine the correlation between the two halves. Here’s how to proceed:
#### **Step-by-Step Guide to Determine Correlation:**
1. **Administer the Test and Split the Items:**
– Administer the complete test to a group of respondents.
– Divide the test items into two halves. Common methods for splitting include:
– **Random Split**: Randomly assign half of the items to each group.
– **Even-Odd Split**: Assign odd-numbered items to one half and even-numbered items to the other.
2. **Score Each Half of the Test:**
– Calculate the total score for each respondent on both halves of the test separately.
– For example:
– If a test has 20 items, compute the sum of scores for the first 10 items (Half 1) and the last 10 items (Half 2) for each respondent.
3. **Compute the Correlation Between the Two Sets of Scores:**
– Use a statistical correlation coefficient, such as **Pearson’s correlation coefficient (r)**, to assess the relationship between the scores from the two halves.
– Formula for Pearson’s \( r \):
\[
r = \frac{\text{Cov}(X, Y)}{\sigma_X \cdot \sigma_Y}
\]
Where:
– \( X \) and \( Y \): Scores from the two halves of the test.
– \( \text{Cov}(X, Y) \): Covariance between scores from Half 1 and Half 2.
– \( \sigma_X \) and \( \sigma_Y \): Standard deviations of the scores from Half 1 and Half 2.
4. **Interpret the Correlation Coefficient:**
– The correlation coefficient will range between -1 and +1:
– A value close to +1 indicates high reliability (both halves measure the same construct consistently).
– A value close to 0 or negative suggests poor reliability (halves do not measure the construct consistently).
5. **Adjust for Split-Half Reliability (Optional):**
– Since splitting the test reduces the number of items in each half, you might underestimate the test’s reliability. Use the **Spearman-Brown formula** to estimate the full test’s reliability:
\[
\text{Reliability}_{\text{full}} = \frac{2 \cdot r}{1 + r}
\]
Where:
– \( r \): Correlation coefficient between the two halves.
#### **Example:**
1. **Test Details**:
– A test consists of 20 items, and responses are collected from 30 individuals.
2. **Split the Test**:
– Split the test into two halves (e.g., odd and even items).
3. **Calculate Scores**:
– Compute the total score for odd items (Half 1) and even items (Half 2) for each individual.
4. **Calculate Correlation**:
– Suppose the Pearson correlation coefficient between the two sets of scores is \( r = 0.78 \).
5. **Adjust for Split-Half Reliability**:
– Apply the Spearman-Brown formula:
\[
\text{Reliability}_{\text{full}} = \frac{2 \cdot 0.78}{1 + 0.78} = \frac{1.56}{1.78} \approx 0.876
\]
This indicates that the test has good reliability.
By determining the correlation between the two halves, you quantify how well the items in one half of the test relate to the items in the other half, providing an estimate of the test’s internal consistency.