Some Interesting Experimental REsults

[Beth]

Hey that's not fair! I did a graph too, but I'm so slow and useless everybody got their's done first. Well I'm posting it anyway 'cos it took me ages so there!

Thats okay. Yours was worth waiting for.

Beth

P.S. I hadn't realized the graphs would be so popular or I'd have posted mine.

 

[alfaniner]

I know this is really obvious, but I'd feel bad if I didn't at least say it.

Beth please do be careful with all these naked flames.

I have the feeling this topic will consists of more flames than any other. Or at least talking about them...

 

[Unnamed]

Unnamed, do you feel that it is valid to compute 2 different linear regressions like that, given that these are not independent trials.

Not really, but it seemed fair to make it that way. By the way, ignore the second linear trend, it's meaningless and I regret putting it there.

I feel that Ashes chart shows that what we are seeing may very well be nothing more than the thermal behavior of the apparatus. Your's gives a strong pschological bias to thinking that there is an effect other than normal thermo behavior.

Personally, I'd make it just like Ashles did, but that would not add anything to the discussion, would it?

A proper control would establish the thermo properties of the apparatus.

I like Genesius' suggestion of alternating trials and controls, because it cancels out this influence.

In any case, both of your graphs are interesting and thought provoking.

And that's the point of making graphs

(edited whitespace only)

 

[Unnamed]

The 9 minute gap is the period during which I was trying to influence the flame. I was working by myself and cannot work with the flame and record data at the same time.

Neither me or Ashles were blaming you for anything. Please don't take it that way. We are saying that the gap in data makes the jump look more impressive than it really is. That's why we made graphs to counterbalance that effect. Every experimental setup has limitations and I agree that the gap was unavoidable.

 

[Gayle]

Beth,

Several times in these forums I have said that Challenge applicants had better toughen up and get used to harsh comments. I mean that. Get strong, stand up, don't snivel.

You have shown both strength and restraint. I strongly support your call for kindness because it's the M.O. I prefer myself. I admire your assertiveness in asking for kindness in a forum where sarcasm and ridicule are more of the norm.

Kramer asked how you were coming with your personal testing and here you are with an answer. That's a pretty good example of getting strong, standing up, and not sniveling. Well done!

Maybe your experiments will succeed and maybe they won't. Regardless of the outcome, you have already succeeded in winning my respect. Good luck and keep posting, please.

Gayle

 

[Startz]

Is it a time trend?

For those who haven't seen Beth's posts from some time back, note that she is studying statistics seriously as well as doing experiments on the paranormal.

With this as an excuse, here's a regression test which controls for a time trend.

Dependent Variable: TEMP
Method: Least Squares
Date: 07/15/05 Time: 21:39
Sample: 1 33
Included observations: 33

Variable Coefficient Std. Error t-Statistic Prob.
C -494.9301 179.2796 -2.760660 0.0097
EFFORT="None"-13.35921 10.14169 -1.317256 0.1977
TIME 1732.921 482.0030 3.595250 0.0011

R-squared 0.802314

As others have suggested, once you control for a time trend the evidence for effort mattering is relatively weak. You'd expect to see evidence this strong about one time out of five.

Beth, if you're interested in this sort of thing, run a regression correcting for serial correlation. You'll find the evidence for the effect completely disappears.
-Dick Startz

 

[Beth]

Beth,

Maybe your experiments will succeed and maybe they won't. Regardless of the outcome, you have already succeeded in winning my respect. Good luck and keep posting, please.

Gayle

Thank you for your encouraging words. I'll keep posting here because I hear enough interesting ideas to make it worth my while. but I'll continue to avoid posting about myself personally due to what I feel is a harsh environment here and my own sensitivities.

I enjoy a good spirited discussion and it's crucial to get a good evaluation of one's ideas; it's when the discussion slips into the personal that I have problems with it. When I'm talking about my own experiments, it's hard to delineate between what is personal and what is not. Hence, I won't be discussing them here anymore.

Beth

 

[Beth]

Re: Is it a time trend?

Beth, if you're interested in this sort of thing, run a regression correcting for serial correlation. You'll find the evidence for the effect completely disappears.
-Dick Startz

Thanks for posting your results. Due to the confounding, I don't think it's appropriate for this data. Of course, it's arguable that the ANOVA I did run isn't appropriate either. At any rate, as I said earlier, this data is not conclusive. However, if you assume that something is occurring, the data support that assumption. To me, that makes it an interesting result. When I have refined my experiment, the confounding should be eliminated and a trend analysis unnecessary.

Beth

 

[Mojo]

The 9 minute gap is the period during which I was trying to influence the flame. I was working by myself and cannot work with the flame and record data at the same time.

So between 8:50 and 9:01, while you were recording the data, you were unable to influence the flame. How do you account for the apparent upwards trend in the temperature during this interval?

ETA: Surely, if the rise in temperature is caused by you directing the flame towards the sensor, and you were only doing this during the 9 minute gap in the data, the sensor should have been heated during those 9 minutes, but started to cool again once you were no longer able to direct the flame towards it.

 

[Beth]

So between 8:50 and 9:01, while you were recording the data, you were unable to influence the flame. How do you account for the apparent upwards trend in the temperature during this interval?

Natural behavior of the flame. I haven't yet estalished a control for comparison because I haven't yet gotten all the equipment I want to work with. When I do, that is something that will have to be done. As Roger point out, you have to know the natural behavior before you can make a comparison.

ETA: Surely, if the rise in temperature is caused by you directing the flame towards the sensor, and you were only doing this during the 9 minute gap in the data, the sensor should have been heated during those 9 minutes, but started to cool again once you were no longer able to direct the flame towards it.

Not necessarily, I was attempting to create a sustained effect because I couldn't check the temp while I was working with it. In addition, a small sustained effect seems to be easier for me to manage than turning the effect on and off, but maybe that just due to confounding.

If I was successful with that attempt, then the results I could expect would be what I got. That's why I find them interesting. However, if the temp had stayed more or less constant throughout the experiment, that would have been an indication of having had no effect.

Beth

 

[Startz]

When I have refined my experiment, the confounding should be eliminated and a trend analysis unnecessary.

Beth:

I agree that refining the experimental design beats using ex post statistics to untangle the results.

I collect interesting data sets for teaching, and maybe for a book I'm writing. Would it be okay if I used the data you posted earlier in the thread? I would identify the author as "Beth," with no last name.
-Dick

 

[deathphoenix]

Beth,

How many sensors do you have? Either situation, I have a suggestion for how you can make your test a little more scientific.

If you have two, set them both up in exactly the same way, and only concentrate on one of the flames (after forty minutes, as you originally tested, if you wish)

If you only have one, set it up in a temperature-stable environment, and have it running for a certain period of time without concentrating on it at all. The next day, set it up in the same environment with the same temperature settings, and concentrate on it (after forty minutes, as you originally tested, if you wish).

Hope this helps.

 

[Beth]

New Question

I ran an experiment earlier tonight. The oil lamp I used turned out to be very smokey. When I was cleaning up, I checked the sensor and sure enough, it was covered with soot. What effect , if any, would this have on the sensor readings? Would it tend to make it higher? or lower? Or would it not make a difference at all?

I'd appreciate any ideas on this and if it might have an effect, do you have any suggestions about how to deal with it. Currently, I'm taping the sensor inside the glass vase, but perhaps I could tape it to the outside? Would that cause any other foreseeable problems?

Thanks for your help and insight.

Beth

 

[deathphoenix]

Re: New Question

I ran an experiment earlier tonight. The oil lamp I used turned out to be very smokey. When I was cleaning up, I checked the sensor and sure enough, it was covered with soot. What effect , if any, would this have on the sensor readings? Would it tend to make it higher? or lower? Or would it not make a difference at all?

I'd appreciate any ideas on this and if it might have an effect, do you have any suggestions about how to deal with it. Currently, I'm taping the sensor inside the glass vase, but perhaps I could tape it to the outside? Would that cause any other foreseeable problems?

Thanks for your help and insight.

Beth

That depends on the type of sensor that you're using (thermocouple, infrared, etc.), but the most common effect of soot covering the sensor is a lower temperature reading than reality because of the insulating effects of the soot buildup. However, since these are pre-preliminary types of testing, I suggest you try both tests (with or without concentrating) and seeing if you have any different readings. Also, you shouldn't be too close to te flame because your presence may also move the flame slightly.

 

[69dodge]

Re: New Question

Originally posted by Beth
Currently, I'm taping the sensor inside the glass vase, but perhaps I could tape it to the outside?

I would try to go in the opposite direction, actually.

Based on the numbers in the first post, I think the sensor is, to a large extent, reading the temperature of the glass, which rises slowly as the small flame gradually heats up the large vase. Can you support the sensor somehow, inside the vase but not touching it? That way, it will read the air temperature, which should change much more rapidly than the glass temperature in response to the changing position of the nearby flame. I'm guessing it's the position of the flame that you hope to affect?

 

[Dave_46]

I think that the soot would make the emissivity of the sensor close to 1, so it would probably increase radiant heat transfer. What it would do to convective transfer I don't know.

Dave

 

[Mojo]

Natural behavior of the flame. I haven't yet estalished a control for comparison because I haven't yet gotten all the equipment I want to work with. When I do, that is something that will have to be done. As Roger point out, you have to know the natural behavior before you can make a comparison.

If you've got the equipment necessary to take any kind of reading, you can easily carry out a control experiment. Simply set up the candle and sensor, exactly as you've done here, and take temperature readings over the expected duration of your actual experiment while making no attempt to influence the flame.

Not necessarily, I was attempting to create a sustained effect because I couldn't check the temp while I was working with it. In addition, a small sustained effect seems to be easier for me to manage than turning the effect on and off, but maybe that just due to confounding.

Or possibly, the fact that your observed "effect" continues after you've stopped trying to influence the flame indicates that it was in fact just the "natural behavior of the flame."

 

[Bodhi Dharma Zen]

A simple protocol would suffice, and account for the variables involved.

1 trial
10 periods control
10 periods aim for sensor
10 periods aim away from sensor
10 periods control

2 trial
10 periods control
10 periods aim away from sensor
10 periods control
10 periods aim for sensor

Repeat ten times, changing the order to test all possibilites

Run some statistics

Present them here.

 

[MRC_Hans]

Re: New Question

I ran an experiment earlier tonight. The oil lamp I used turned out to be very smokey. When I was cleaning up, I checked the sensor and sure enough, it was covered with soot. What effect , if any, would this have on the sensor readings? Would it tend to make it higher? or lower? Or would it not make a difference at all?

I'd appreciate any ideas on this and if it might have an effect, do you have any suggestions about how to deal with it. Currently, I'm taping the sensor inside the glass vase, but perhaps I could tape it to the outside? Would that cause any other foreseeable problems?

Thanks for your help and insight.

Beth

Taping it to the outside would make it sensitive to outside influence, like your breath, and indeed your presense.

Ideally, you should have two sensors, on on each side of the flame, and once you have balanced them, you should measure the difference between them. I can design the circuitry needed. This would eliminate most of the heat-up problems and other problems of varying intensity of the flame.

Soot problems can be eliminated by using an alcohol flame.

Finally you should use a randomized scheme, and a blinded read-out would be preferable (would require either a datalogger or a helper).

Hans

 

[roger]

Beth,

As you may recall, I suggest the thermocouple apparatus to you some time ago, after trying it out in my own kitchen (I have one readily available to me). My set up was quite simple, a candle put inside a coffee cup, and the thermocouple held by hand, suspended in the air inside the cup.

I noticed very significant fluctuations in the temperature over time. Significant meaning fluctuations of +/- 10-20 degrees F over a few seconds. So it might read, in 2 second intervals: 130 134 142 130 128 120 145 165. (I just made that up).

There are a few reasons for that, I hypothesize. One, the thermocouple I have is very sensitive, as is normal for these devices. Second, hand holding adds variations. But most important, I suspect air currents play a large role. Burning a candle generates complex, perhaps even chaotic air patterns, because as the candle burns it burns oxygen, causing new air to be drawn in. And of course the high heat sets up convection currents.

The upshot is that I am not particularly surprised by your data, and don't trust it without knowing more. You gave readings once per minute. With my thermocouple set up, if I chose to, I could easily pick readings that showed an upward trend, a downward trend, or no change, just by varying when I took the reading by a few seconds. I'm not accusing you of cooking the results, but I'm sure you can see how unconscious bias might lead to the same result.

As others have suggested, you can already do a control trial. Do a burn for 20 minutes, and record the data. Shut down the system, and allow all temperatures to stabilize back to room temp. Do this several times, and you should start to see some patterns. If the data is consistant, then you are ready to try your test. If it isn't, then you need to find out why. Air currents? Differing candle height? Candle/wick variations? Soot buildup? Etc.

Once you have a reproducable control procedure which is very sensitive to changes in temperature, then you can proceed. Until then, I think you will end up chasing phantom effects when trying to control the candle with your mind. While what I wrote above may sound time consuming, it'll take a lot less time then running tests that aren't giving meaningful data, that may in fact give you misleading data (example: you may think you are moving the candle towards the flame, but you are really moving it away, but this is obscured by the error in the apparatus).

I'm sure you thought of a lot of this already, but I hope there was something in there that was helpful.