representative sample!

[Andy_Hollywood]

hello

I have to test a number of people typing:
"The quick brown fox jumped over the lazy dog"

and then compare there input results, like how long they hold down a key for and how long between keystrokes.

The problem i have is how many times should i get them to type the phrase, and how many people should i run the test on!?

Hope some one can help, cheers in advance, and if you ned me to elaborate then let me know

Andy

[Guv]

It is my guess that you want a lot of people to type the sentence once. This would result in a sample of typing skills. I would expect people to improve or at least change their typing actions slightly if asked to type the same sentence more than once.

BTW: That sentence is known to contain all 26 letters of the English alphabet. It should be typed with the second the replaced by a, making it slightly shorter.

Another (and shorter) sentence containing all 26 letters is.

Pack my box with five dozen liquor jugs.

I have never seen a shorter sentence with all 26 letters except for poorly constructed sentences, or ones containing proper names, or ones containing incorrect English words, or ones that did not make good sense.

[Andy_Hollywood]

thanks very much for the suggested sentance, i will include it!

As for the number of people to test, i just have to try and get an average for each person. So i can then compare the group.

I found some notes on a similair test that was carried out and they tested 13 people each typeing the sentance 32 times


how would they have arrived at these number? or are they just plucked out of the air?

[Guv]

Sounds to me like they just guessed at 32 repititions.

It seems to me that the data becomes meanless after a person types the same sentence 5-10 times.

Some people are likely to get bored and slow down or make mistakes (or both) due to day dreaming or aggrevation. Others will surely get faster due to practicing on the same sentence.

Surely something would be expected to change after typing the same sentence 10 or more times. I might decide to quit or not even start if not being paid.

[Andy_Hollywood]

I see your point, so do you think that if i get a person to type the sentance 10 times i will have enough information to find out there average so to speak?

I will be generating logs like:

KeyDOWN:70,Sat Dec 29 21:29:20.252 2001
KeyUP:70,Sat Dec 29 21:29:20.362 2001
KeyDOWN:71,Sat Dec 29 21:29:20.382 2001
KeyDOWN:74,Sat Dec 29 21:29:20.382 2001


and then looking at the amount of time between keystrokes and the time the key is down for etc..

but then how many people to test on? the more the better? or is there a number that can be representative?

[Guv]

It is my understanding that the best sample is the largest possible sample with the members chosen randomly. An alternative is to decide on various representative groups and weigh the results.

An example of the latter would be to choose two or three people from each of 5 or 6 age groups. Then weight the results by the population statistics for the 5-6 groups.

The manner of choosing a sample depends a lot on what you are trying to measure. An obvious example: If you want to market a product to male college students, pick some male college students for a test of your advertising copy.

I do not think one size fits all here.

[Guv]

A bad sampling technique can make a mess of a statistical project.

A classic example was a poll done during the great depression (some time between 1929 & 1936). The pollsters picked random people from telephone directories. At that time, only fairly affluent people had telephones, resulting in a biased sample and erroneous results.

Another way to get a biased sample is to advertise for volunteers.

[Andy_Hollywood]

I am doing this study into Biometrics using keystroke dynamics and have to look at discrimination factors between people, hence having to look at loads of peoples keystroke timings.

When i finally get round to sampling people i will have to advertise, or email the undergraduates on my course.

In such a study do you think it is possible to get a bias as i will be looking at characteristics that are unique to a person?!?!?

[Guv]

You will get data unique to the class of people you sample. At least try not to mention the task being tested.

If you mention that the task is typing, you are likely to get only those who are confident of their typing skills, an obviously biased sample. Contacting people by email is likely to result in getting a bias in favor of the better typers. Wouldn't you expect the people who do not have an email address to be the poorer typists?

It seems to me that you are going to be sampling college students, not the general population. If this is reasonable, okay. Just remember that the general population includes a healthy percentage of people who have never used a keyboard. I would expect over 95% of college students to have had experience with Word Processors, email, et cetera, while I would guess that less than 40% of the general population has such experience. In an underdeveloped country, probably less than 10-20% have even seen a keyboard.

If you recognize that your sample is not representative of the general population, there is no harm. To be professional, you should mention that it is not representative, and probably should describe your results as reflecting the typing habits of colleger students.

[Andy_Hollywood]

a very good point, i shall make a note.

so if i point out the faults with my sample, and my reasoning behind them i should be ok?!

thank for your help

Andy