Rasch analysis

The Rasch analysis report produces a series of tables and visualisations to help analysts interpret the underlying psychometric properties of the test.

As a 1-parameter model, it asserts that the probability of endorsing a correct answer is a product of the relationship between the test-taker's ability and the difficulty of the item.

If you are interested in learning more about the Rasch model (including its assumptions), the Wikipedia page is a good place to start: https://en.wikipedia.org/wiki/Rasch_model

Running the Rasch analysis report

To run the Rasch analysis report:

Step 1 of 4

Navigate to your exam then Set standard and View reports.
Step 2 of 4

Click Create new report and Rasch model report.
Step 3 of 4

Select the date for which you wish to run the report. You can also optionally give the report a name. Select the items you wish to include in the report (SHIFT+select or CMD/CTRL+select for multi-selection). Alternatively do not select any items in the list to include everything.

Click Save to commit.
Step 4 of 4

Click Preview to view the output.

You can also export a copy of this report to PDF.

Report output

The expandable sections below give an overview of the different sections of the report output. For further in-depth guidance, please see here: https://www.winsteps.com/winman/misfitdiagnosis.htm

General terminology and measures

Metric	Description
Infit mean-square (Infit MNSQ)	Inlier-sensitive fit statistic derived from the chi-square statistic.
Infit z-standardised (Infit MNSQ)	Inlier-sensitive fit statistic standardised to z-score
Outfit mean-square (Outfit MNSQ)	Outlier-sensitive fit statistic derived from the chi-square statistic.
Outfit z-standardised (Outfit (ZSTD)	Outlier-sensitive fit statistic standardised to z-score
Point measure correlation (PTMEA)	Correlation between the measure and the overall model.

Colour coding

Colour	Description
Red	Value is significantly concerning on the high side
Yellow	Value is concerning on the high side
Blue	Value is concerning on the low side
Purple	Value is significantly concerning on the low side
Green (in summary table only)	Value is within expected range

The Rasch analysis table gives a summary overview of the model and all statistics that follow in the report. It tests against expected values and indicates whether or not the tests are met.

In the example shown above:

Column	Description
Psychometric property	The property being measured
Statistic	The statistic being used to measure the property
Count	Raw data values. In the first row we can see that 1 of the 29 items has an Outfit-MNSQ value >=2.0 in the Item parameters and fit statistics table, hence 1/29 is the value.
Test	The expected value against which the result is being compared.
Result	The result statistic, colour coded to indicate the outcome of the comparison against the expected value. In row 1, the result is 1/29 = 3.45% which is greater than the 0% expected value and colour coded in yellow.

The item summary provides a simple overview and summary statistics of the detailed values presented in the Item parameters and fit statistics table.
An abridged version of the table is shown in the image. The full table contains one row for every item in the exam and shows detailed fit statistics for each.

In this case the Measure attribute is the item difficulty.
The person summary provides a simple overview and summary statistics of the detailed values presented in the Person parameters and fit statistics table.
An abridged version of the table is shown in the image. The full table contains one row for every candidate in the exam and shows detailed fit statistics for each.

In this case the Measure is the person's ability.
The standardised residual variance table shows the amount of variance in the data that can be explained by the model. In this case around 30% of the variance is explained, leaving the inverse value (around 70%) unexplained.

The contrasts estimate other sources of variance present in the model.
This graph plots the contrasts information given in the Item standardised residual variance table. Note contrast 1 at just below 14% which drops to just below 8% for contrast 2, and so on.
This chart plots the first two residual components of the model. Ideally, the distribution of points should be random. Any clustering may indicate that the residual is influencing the results.
This chart plots and visualises the distribution of the estimated item difficulties and person abilities against one another. In the example shown, the person frequency of candidates higher on the ability scale contrasts with the items which are generally lower on the difficulty scale. Essentially our items are not discriminating well at the higher ability levels.
This plot shows us where on the ability scale the exam is providing the most information. In this case the test gives us the most information about candidates just below 0 on the ability scale. Inversely, the standard error of measurement (SEM) is therefore lower around the same point on the ability scale and higher at the extremes.