Questions & explanations
1. Compare the information function of a Rasch item to a 3PL item with high discrimination.
The information function shows how much an item contributes to measuring ability at different levels. For a Rasch item, the information is highest when ability equals difficulty, and it has a symmetric bell shape. For a 3PL item with high discrimination, the information peak is narrower and taller, meaning it provides more precise measurement at a specific ability range. However, the guessing parameter reduces information at low ability levels. So a 3PL item with high discrimination is very informative for a narrow range but less useful outside that range. In contrast, a Rasch item provides moderate information over a wider range. Test designers use these functions to select items that maximize information at the target ability.
2. How does the 3PL model differ from the Rasch model in handling guessing?
The 3PL model includes a guessing parameter (often called the pseudo-guessing parameter) that lowers the probability of a correct answer for low-ability students to a non-zero floor. In the Rasch model, the probability approaches zero as ability decreases. The 3PL model also allows items to have different discrimination parameters, meaning some items are better at distinguishing between students of similar ability. The Rasch model assumes all items discriminate equally. Because of these extra parameters, the 3PL model is more flexible but requires more data to estimate accurately. The Rasch model is simpler and often preferred for its mathematical properties like sufficient statistics.
3. Compare corporate e-learning with traditional in-person training in terms of flexibility.
Corporate e-learning is much more flexible than in-person training. Employees can access materials anytime, anywhere, on any device, fitting learning around their work schedule. In-person training usually happens at a fixed time and place, requiring travel and time away from work. E-learning allows learners to pause, rewind, and review content as needed, which helps understanding. In-person training offers immediate face-to-face interaction and hands-on practice, which e-learning can simulate but not fully replace. However, e-learning is more scalable: the same course can reach thousands of employees at once. Many companies now use a mix of both methods for best results.
4. What does the Rasch model assume about item difficulty and person ability?
The Rasch model assumes that the probability of a correct answer depends only on the person's ability and the item's difficulty. It uses a simple logistic function where the difference between ability and difficulty determines the chance of success. The model assumes all items have the same discrimination (slope) and no guessing. This makes it a 1-parameter logistic (1PL) model. The 3-parameter logistic (3PL) model adds two more parameters: discrimination and guessing. Discrimination allows items to vary in how well they separate high and low ability students. The guessing parameter accounts for the chance of a correct answer by random guessing on multiple-choice items.
5. If a test uses the Rasch model, what does it mean when an item has a high difficulty parameter?
A high difficulty parameter means the item is harder, so only students with high ability are likely to answer it correctly. In the Rasch model, the difficulty is on the same scale as ability. For example, if a student's ability equals the item difficulty, they have a 50% chance of answering correctly. A high difficulty item requires a higher ability to reach that 50% probability. Test designers use these parameters to match items to student ability levels. In adaptive testing, easier items are given first, then harder ones as ability is estimated. The Rasch model's simple relationship makes it easy to compare students and items on the same ruler.
6. Compare blended learning with traditional classroom for indigenous students.
Traditional classroom learning usually uses the same books and methods for all students, often in a national language. Blended learning for indigenous students adds online parts that can be customized to their culture and language. For example, instead of only reading a textbook about farming, they might watch a video of a relative explaining local plants. Blended learning also allows students to learn outside school hours, which helps if they have chores or ceremonies. However, it needs internet access and devices, which some communities lack. The best approach uses both methods together to respect traditions while teaching modern skills.
7. Compare maximum information and shadow-test item selection in terms of constraint handling.
Maximum information item selection does not naturally handle constraints like content balance or item exposure limits. It may repeatedly pick items from the same content area or overexpose a few high-information items. To address this, modifications like randomesque or exposure control are added. In contrast, the shadow-test method explicitly incorporates constraints into the selection process. It ensures that every selected item is part of a feasible full test that meets all constraints. This makes shadow-test more suitable for high-stakes tests where content coverage is critical. However, shadow-test is computationally more intensive.
8. Give an example of how the rule space model would classify a student who answered 3 out of 5 fraction problems correctly.
Suppose the test has 5 fraction items testing skills like 'simplify', 'add', 'subtract'. The rule space contains all possible mastery patterns (e.g., has 'simplify' but not 'add'). For each pattern, an ideal answer vector is computed (which items would be correct if the student had those skills). The student's actual answers are compared to each ideal vector. The model finds the pattern with the smallest difference, considering that a student might slip on a known skill or guess correctly on an unknown one. So if the student missed an 'add' item but got a 'simplify' item, the model might classify them as having 'simplify' but not 'add'.
9. Compare the Community of Inquiry framework to other design models for blended courses.
The Community of Inquiry framework says that good blended learning needs three things: teaching presence, social presence, and cognitive presence. Teaching presence means the teacher helps guide learning; social presence means students feel connected; cognitive presence means they think deeply. Other models, like the SAMR model, focus more on how technology changes tasks from simple to more complex. Both are useful, but the Community of Inquiry is better for designing whole courses because it looks at how people interact. SAMR is simpler for choosing specific tools. Blended course designers often use a mix of both to plan lessons.
10. Compare the information provided by a dichotomous item and a polytomous item with three categories.
A dichotomous item provides information about whether a person is above or below a single threshold. A polytomous item with three categories provides information about two thresholds, so it can distinguish between three levels of ability. This often results in more total information across a wider ability range. For example, a partial-credit item can tell if a student has low, medium, or high ability, while a right/wrong item only tells low or high. However, polytomous items require more data to estimate parameters. In adaptive testing, polytomous items can be very efficient because they give finer-grained information per item.
11. What is professional development for blended learning?
Professional development for blended learning is training that helps teachers learn how to teach with both online and in-person methods. It can be workshops, online courses, or coaching where teachers practice using technology and designing lessons. The goal is to help teachers feel confident in choosing the right tools and activities for their students. Good professional development includes hands-on practice, examples from real classrooms, and time to plan. It also teaches how to manage a blended class, give feedback online, and keep students engaged. Without this training, teachers may struggle to use blended learning well.
12. Compare the complexity of estimating a MIRT model versus a unidimensional IRT model.
Estimating a MIRT model is more complex because it involves multiple dimensions and their correlations. The number of parameters increases: each item has a discrimination parameter for each dimension it loads on. The likelihood function is multidimensional, requiring numerical integration over several dimensions, which is computationally intensive. Unidimensional IRT only integrates over one dimension. MIRT also requires larger sample sizes to get stable estimates, especially with many dimensions. However, software like flexMIRT or mirt in R can handle these models. The trade-off is more realistic modeling versus simplicity.