Please Help - How Do I Prove Hypothesis?

Sci

New Member
#1
Hi,

I am in 6th grade, trying to do an analysis of data for a science fair project. Sorry, but I am only at Algebra 1, I do not have any background in statistics. I have been reading and trying to understand what test(s) to use, but with no luck.

My experiment is can a blindfolded person sense the distance of an object by electric shock to the tongue? As the object gets closer, the shock gets stronger. It is not a timed race, it is just pass (1) or fail (0). The null hypothesis would be all failures, that is if the device did not work, the person could not see the object because he or she was blindfolded. The only possible results are 1 or 0. I had 17 people each try 4 different times to do 4 different object detection tests.

Here is a sample of my data (I have 17 people, not just two people):

Person A Test One: 1, 1, 1, 1 Test Two: 1, 1, 1, 1 Test Three: 1, 1, 1, 1
Test Four: 1, 1, 1, 1

Person B Test One: 1, 0, 1, 1 Test Two: 1, 1, 1, 1 Test Three: 1, 1, 1, 1
Test Four: 1, 1, 1, 1

The data seems to show that the device works, if it did not, then the data above would be all 0s (all fails).

But, here is my question: How do I statistically test my data to show that my device works, that the null hypothesis is rejected? I have 17 different people, so I cannot use a paired T Test, so I am totally lost. Can someone please help me, what statistics can I use to show that my device is successful, that the results are not random chance? Do I show results statistically by each test, or all tests, by each person, or as what?

Thank you
 

Sci

New Member
#3
The object is a video camera and ultra sonic pinger (proximity detector) connected to an interface which then is connected to a3.5 cm square array of electrodes held in the mouth, on the tongue. The person is blindfolded and cannot see anything. The person has to walk through a maze and not bump into a 2m x 2m cardboard object. For safety (I did not want people to fall and get hurt while blindfolded) only I tried to do the maze with the device turned OFF and also being blindfolded. I knocked over the object every time, since my friend placed the objects in random places for each test. When the 17 subjects used the device, they seldom knocked over anything, they were able to perceive the distance of the objects.

So it sure seems that the device works, but that is just my opinion. How do I prove it statistically? And, I am sorry, I really dont know anything about statistics. I did try by myself to do this, but all I could find was an anova or t-terst, and I do not know if they are the correct way to test this. Sorry that I know so little, I wont be taking statistics until 11th grade.
 
#4
Good project, Sci! Is this by any chance for the Intel science fair? I was a judge at my local one last year, and I think this would be an excellent project to submit! I think you'll do just fine with Algebra I.

Dason asked a really good question. Do you understand why it's important? And can you explain why? (Sometimes understanding the reason and being able to explain it out loud are two different things.) If I'm your judge, I'm going to ask you that also. And I'm going to expect you to be able to explain it.

I would re-think your null hypothesis. That might be an OK way to explain it, but it isn't really a null hypothesis. For your presentation, you need to come up with a real null hypothesis and a real alternative hypothesis. They should be one sentence each. I like to use the words "there is no relationship" or "subjects cannot" or something like that for a null hypothesis and the opposite for the alternative hypothesis. Short is good. Short fits on your presentation board.

One caution: when writing up your results, make sure you really understand the meaning of "statistical significance." I judged a number of entries last year where the student properly reported a significance value of 0.1 or more, but then talked about their results anyway and drew conclusions as if their data really meant anything. If you want to have a shot at winning your category, you need to understand that sentence and understand why those students made a mistake. The student who won my category was the only one who didn't make that mistake.
 

Sci

New Member
#5
I hope to enter the Intel science fair, but that will be decided after I participate in my school's competition. i never did a science fair project before.

I am sorry, I am lost with the null hypothesis issues you raised. My hypothesis is that if the "see in the dark" device is turned on, a person can detect objects in the dark. The null would be its compliment, if the device is turned off, then a person cannot detect an object in the dark. for safety, I am not running the null, because people may fall and get hurt, so the null is assumed that with the device turned off, then a person could not "see" an object in the dark. Am I missing something? I want to show from the data that with my "see in the dark" device turned on, people can actually see inthe dark, as shown in the tests done by 17 people in avoiding cardboard obsticles (not trying to force the results, but in 17 subjects doing 4 trials of test 4 seperate times, there were failures to see only 3 times, so the "see in the dark" device does, on its face seem to really work - but what test can say this instead of just a kid's opinion?)

I am not a statistician, so I wont lie to a judge, there are so many things I do not know, but I will try to find out and I will always do my best. I would not use 1% as my significance level or level of certainty, I would not reject the null without a 5% significance level.

My issue is what tests to apply? I would love to learn statistics, but I am less thn 2 mounths into algebra 1, and I am not at the level to understand such complicated calculations. I can, I hope, focus on doing a specific test, or tests, if I am pointed in the right direction. But, I am lost, rudderless in the ocean, unable to move forward : (
 
Last edited:

Dason

Ambassador to the humans
#6
The problem with not running the control part of the experiment (where they don't have the object) is that you have nothing to compare to.

It would be like if I told you "My friend Steve is better at free throws than Dan because Steve has a free throw percentage of 80%". You don't know anything about Dan so me telling you that Steve's percentage is 80% doesn't tell you if Dan is better than Steve or not.

One of the most important things to include in a scientific experiment is a control group so that you can make some sort of inference regarding the treatment.
 

Sci

New Member
#7
I am sorry, I have been sooooooooooooo dense! My whole set of trials are not saying anything, other than 17 differemnt people had this result.

If I re do my experiment, and I have each person do the 4 trials with the "see in the dark" device turned on and then again with it turned off (I will make sure it is safe, and that everyone takes small steps) then I will have results for 17 people walking through 4 different mazes each, and doing each maze 4 times with the device turned on, and 4 times with it turned off, then I will have something to compare. BUT THIS IS WHERE I AM DENSE AGAIN - HOW DO I THEN STATISTICALLY TEST THE DATA, WHAT TESTS DO I RUN TO SEE IF THE RESULTS SHOW WITH SIGNIFICANCE THAT THE DEVICE WORKS?
 

Dason

Ambassador to the humans
#8
I can think of a model that might be appropriate for that situation but to be honest it's probably too complex for you to reasonably run and analyze. However, for a simplified case of your situation McNemar's test sounds like it could be appropriate.
 

Sci

New Member
#9
Thank you. I will start researching that test. And, thank you again and Berley too, my whole experiement setup was flawed, my results were worthless, even though I worked very hard on them. It would have been a meaningless project, but thanks to you, I can see how to re-run my experiements and then start testing the results. THANK YOU - I am going to research that test today, so that I can do a good job using it!
 

Sci

New Member
#10
PS If there is any other test, such as an anova or something that I should also consider, please tell me, and I will look it up and figure out how to run it also. I do have excel to do my calculations, but I still will need to learn to know what those numbers are telling me. I am open to any suggestions, and ready to research and do the work! THANK YOU : )
 
#11
Good thoughts so far, Sci. You're not dense -- you're right on the verge of putting everything together. I agree with Dason that McNemar's test is a good one for you to use if you have a control group. This is where your null hypothesis is going to come into play. Yes, your null and alternative hypotheses are opposites of each other.

I might suggest you consider a null hypothesis like:
Subjects using the sensor device do not perform better than subjects not using the sensor device.

That makes your alternative hypothesis:
Subject using the sensor device perform better than subjects not using the sensor device.

See why Dason's question was important? Now, given what Dason told you about the test, can you see how to use McNemar's test to see if your results are statistically significant? And if they are statisically significant, can you see what conclusion you can draw?

Since it looks like you are going to re-run your experiment, I'll throw out one other idea. If there is only one object in the room, maybe your test would be better if you set it up so the subject could find the object instead of avoid it. So it might have a real-world application for blind people. They could wear a sensor and tag important objects in their home and then a vibrating (not shocking) sensor could help them find those important objects. Just something for you to think about. Science for it's own sake is fine. But science that helps improve people's lives is better.
 

Sci

New Member
#12
Hello Berley,

Thank you for your kind words and your help! Fantastic ideas for phrasing my hypothesis and null - I can see now how your suggestion helps to zero in and focus on the main point of comparison.

I took a quick look at McNemar's test, I see how it puts everything together and shows yes or no, does this really work. however, I have much more to study about this test, I want to do more tham press some calculator bttons, I want to understand the reasons behind the test and what the analysis is saying. Since you both were so kind to point me in the correct direction, I need to do my part and not just use the test but understand it! I do see now that the paired t-test does not work for me because I have 4 different mazes, this I did not understand until I looked at the McNemar's test.

I am going to re run my experiment, you both saved me from being a fool - my old results were meaningless, they said nothing. Sure, I worked hard, but I made an error. Now that I know the error, I must fix the error and re do my work. Otherwise I am knowingly doing false science, that would be horrible. Hmmm, what a wonderful idea, instead of avoiding the object, find and tag it. Yes, I will try that. I did try a vibrator, but it did not work very well. The shock on the tongue really catches a person's attention, I know that is why I chose it, it got my attention every time!

THANK YOU AGAIN!!!!
 
#13
You're very welcome!

One other thought: keep all your old results from the first round and print out this discussion chain. Put all of them in your research journal. They are good examples of the process you've used to start to analyze your data and design (well, redesign) your experiment. Plus, they help document that you actually did the work and that someone else didn't just tell you the answer.

Good luck and keep us posted! Post again on this thread, or post a note here that says you started a new thread.

(And if you haven't watched the Mythbusters pirate episode about eyepatches and navigating a maze in the dark or the one episode about driving with your headlights off in the dark -- the one with the go karts -- you should try to watch them both. They may not be directly applicable to your experiment, but you might get some ideas about constructing mazes when the subject can't see what's around them.)