Experimental Design B/C

[sciolyscorpio95]

That is seriously messed up...there's a reason DV stands for Dependent Variable because guess what? It DEPENDS on the independent variable.

Edit: sciolyscorpio, this was at PA States? Wow...I expect better from such a generally well-run state.

I know, it was really strange! It wouldnt bother me if it was just at a couple invitationals, but states is pretty serious. i mean, thats where it actually counts. we found it really hard to believe, but thats the sad truth.

[Teal]

Guys, I'm unsure of what to do for Analysis and conclusion. What's the difference ...?
I'm also a bit wary of my Standard of comparison. For example, if we're measuring the effect surface area has on friction (yeah, I just made that up), and we're cutting out different squares of paper, what would be the standard?

Oh, and graphs. I have no idea what to do with graphs.

Typically, I'm just the procedure/materials/first half of the lab person, but I want to do well at state.

[Phenylethylamine]

Guys, I'm unsure of what to do for Analysis and conclusion. What's the difference ...?

For the analysis, you're talking about trends in the data, while for the conclusion, you're talking about what's probably causing those trends.

In effect, this means that your analysis is basically a more in-depth restatement of your statistics. The rubric says something about mentioning every data point, but they don't literally mean "The first trial of this level of the IV was 4.26 seconds, the second trial was 4.84 seconds, and the third trial was 4.61 seconds"; it's more like "All trials at this level of the IV were within one standard deviation of the mean." If one of the trials wasn't within one standard deviation of the mean, you would talk about how it's an outlier. You also need to talk about the overall trends (e.g., "There appears to be a positive linear correlation between [IV] and [DV]"); this is where you'd bring in R-squared values or similar in C Division. The analysis is strictly looking at statistics and trends in your results. This is not where you speculate on why you got the results you're seeing.

The conclusion is where you first address your hypothesis, then show off the fact that you know science. The first thing you have to do is talk about your hypothesis in terms of your results (e.g.,"Our hypothesis that [DV] would increase as [IV] increased is supported by our data, because we found a positive linear correlation between [IV] and [DV]"). Then you can speculate on why you got the results you're seeing (e.g., "Our hypothesis that effervescing time of a mixture of acetic acid and sodium carbonate would increase as the volume of sodium carbonate in the mixture increased is supported by our data, because [correlation blah blah]. This occurs because the sodium carbonate is the limiting reactant, so increasing the amount of sodium carbonate allows a greater amount of product- in this case carbon dioxide- to be produced. Because the reaction rate remains roughly constant, this translates to a longer effervescing time." For this example, it would be a good idea in C Division to pull out some stoichiometry as well and show that Na2CO3 was, in fact, the limiting reactant).

[scifipi]

Oh, and graphs. I have no idea what to do with graphs.

Graphs really don't have to be too complicated. Record the numerical data from your trials (preferably 2-4 of them) and plot them on a line graph or have bars for a bar graph. When doing each variable a certain amount of times, use different colors and a key to distinguish between variables. Remember to label your x and y axis, and to title your bar graph. You can also do double, triple, or even quadruple bar graphs (2, 3, or for in per clump).

[Phenylethylamine]

Oh, and graphs. I have no idea what to do with graphs.

Graphs really don't have to be too complicated. Record the numerical data from your trials (preferably 2-4 of them) and plot them on a line graph or have bars for a bar graph. When doing each variable a certain amount of times, use different colors and a key to distinguish between variables. Remember to label your x and y axis, and to title your bar graph. You can also do double, triple, or even quadruple bar graphs (2, 3, or for in per clump).

For C Division, you need to do a scatter plot- not a bar graph- you need error bars on your data points, and you don't connect them with a line; you do a line of best fit.

[quizbowl]

Oh, and graphs. I have no idea what to do with graphs.

Graphs really don't have to be too complicated. Record the numerical data from your trials (preferably 2-4 of them) and plot them on a line graph or have bars for a bar graph. When doing each variable a certain amount of times, use different colors and a key to distinguish between variables. Remember to label your x and y axis, and to title your bar graph. You can also do double, triple, or even quadruple bar graphs (2, 3, or for in per clump).

For C Division, you need to do a scatter plot- not a bar graph- you need error bars on your data points, and you don't connect them with a line; you do a line of best fit.

What do we do if the relationship is nonlinear? Do we just write in analysis as the reason why the r squared values are so low is that reason?

[Phenylethylamine]

What do we do if the relationship is nonlinear? Do we just write in analysis as the reason why the r squared values are so low is that reason?

Yeah, you should still have a best fit line; ideally, you'd be able to fit some other curve, but good luck doing that on your TI-30. Like you said, you would then mention in the analysis that because the r-squared values (or whatever other measure you're using) are so low, your data may be nonlinear.

In most cases, either you'll be doing an experiment that should give linear results, or it'll be on such a small scale that the parabola/exponential/whatever actually looks linear. There are some exceptions, but mostly you can just treat your data as linear.

[The only real exception I can think of off the top of my head is if they give you M&Ms and ask you to model radioactive decay. I doubt they'd do that, since it's so simple, but if they did- at an Invitational or something- you should treat your data as exponential and not try to fit a line.]

[jayadh]

So for states does anyone have an idea of what the experiment might be (C division and B) please and thank ya!

[nejanimb]

What do we do if the relationship is nonlinear? Do we just write in analysis as the reason why the r squared values are so low is that reason?

Yeah, you should still have a best fit line; ideally, you'd be able to fit some other curve, but good luck doing that on your TI-30. Like you said, you would then mention in the analysis that because the r-squared values (or whatever other measure you're using) are so low, your data may be nonlinear.

In most cases, either you'll be doing an experiment that should give linear results, or it'll be on such a small scale that the parabola/exponential/whatever actually looks linear. There are some exceptions, but mostly you can just treat your data as linear.

[The only real exception I can think of off the top of my head is if they give you M&Ms and ask you to model radioactive decay. I doubt they'd do that, since it's so simple, but if they did- at an Invitational or something- you should treat your data as exponential and not try to fit a line.]

Doing data transformations on a calculator that's legal for this event is really not all that complicated. There are tons of instances where you'll do experiments with non-linear results: inverse relationships (pressure-volume labs with a gas, for instance), square root (the most basic of all: pendulum labs), inverse square (like the static electricity lab at nationals two years ago), etc. So many things aren't linear! Usually, you'll know it in advance and your hypothesis should predict that non-linear relationships, and then your data and analysis section should have transformations (not too difficult). A thorough lab will consider non-linear possibilities if you didn't know the relationship beforehand and just predicted linear but get results that are close but not quite - parabolas that fit, for example. Do a rough sketch of a residual plot to check for curves, and do a transformation and compare R^2 values.

[JustDroobles]

What kind of statistics do you guys use for your "other relevant statistic"?