WEBVTT

00:00:00.390 --> 00:00:08.360 position:50% align:middle
Here, let's look at a hypothetical example of someone
taking the exam, just to illustrate how all this work.

00:00:08.360 --> 00:00:14.800 position:50% align:middle
So here you see the x-axis, the horizontal axis,
that represents the order of the items.

00:00:14.800 --> 00:00:24.734 position:50% align:middle
So we have the 1st item to the 15th item,
and then the y-axis here is the difficulty of the item.

00:00:24.734 --> 00:00:30.413 position:50% align:middle
So for the first item that was administered to
this particular person, it's a little bit under the

00:00:30.413 --> 00:00:35.913 position:50% align:middle
passing standard. The red line is
how much you need to have to pass.

00:00:35.913 --> 00:00:40.433 position:50% align:middle
Person answers this item, she got it correct.

00:00:40.433 --> 00:00:48.249 position:50% align:middle
From that, we got an ability estimate,
and the line going through the ability estimate is our

00:00:48.249 --> 00:00:50.597 position:50% align:middle
95% confidence interval.

00:00:50.597 --> 00:00:56.712 position:50% align:middle
What it means is that, at this point, we are
95% certain that the person's true ability

00:00:56.712 --> 00:00:59.974 position:50% align:middle
estimate is somewhere in that range.

00:00:59.974 --> 00:01:07.314 position:50% align:middle
And from this ability estimate, we get
a little bit information about this candidate,

00:01:07.314 --> 00:01:12.156 position:50% align:middle
and then we use that
information to select our next item.

00:01:12.156 --> 00:01:15.236 position:50% align:middle
Person answered the next item correctly.

00:01:15.236 --> 00:01:19.536 position:50% align:middle
From that we got a new ability estimate.

00:01:19.536 --> 00:01:24.056 position:50% align:middle
We administered the third item. She got it wrong.

00:01:24.056 --> 00:01:29.801 position:50% align:middle
So then the subsequent
ability estimate's a bit lower.

00:01:29.801 --> 00:01:32.951 position:50% align:middle
Next item, she got it wrong again.

00:01:32.951 --> 00:01:37.011 position:50% align:middle
New ability estimate is lower than the one before.

00:01:37.011 --> 00:01:41.056 position:50% align:middle
Next item, so on and so forth.

00:01:44.161 --> 00:01:51.318 position:50% align:middle
So for every answer that a candidate gives
on an exam, we use that information to estimate

00:01:51.318 --> 00:01:52.935 position:50% align:middle
that person's ability.

00:01:56.731 --> 00:02:03.890 position:50% align:middle
Another thing that I would like to point out is that
you see these, 95% confidence interval. As the test

00:02:03.890 --> 00:02:11.450 position:50% align:middle
progresses and we get more information about
this person, this 95% confidence interval will get

00:02:11.450 --> 00:02:12.944 position:50% align:middle
narrower and narrower.

00:02:12.944 --> 00:02:16.470 position:50% align:middle
We get our estimates getting more and more precise.

00:02:16.470 --> 00:02:23.688 position:50% align:middle
So you'll see later, by the time we administer
the whole test, the error is very small.