Any Interesting Statistics Studies?

Question

A good well developed problem and a nice way to analyze the data and come to a conclusion about the problem.

ex. effect of ..... on .....

any good topics or titles?

Answer 1

the previous answer does mention a nice idea for a data set. but be careful of how you interpret a statistically significant relation between the djia and oil price.

there is a famous data set giving numbers of babies born and number of storks for the different counties of england for a certain year. if you compute the correlation between number of babies and number of storks, it is positive and significantly different from zero. so do we conclude that storks bring babies?

the analysis of the djia and oil price may show that they move together... but it doesn't follow that one causes the other to move. both may be influenced by other variables that cause them both to rise or fall together.

in the case of the storks and babies, that third variable is the size of the county. larger counties have larger numbers of births and also larger numbers of storks.

one other point - the djia and oil prices are examples of time series. that usually means successive values are correlated with previous ones. methods for analyzing time-series data usually take into account this so-called autocorrelation between the successive values of each time series. you might be opening a can of worms in working with such data.

if your timeframe for doing a study allows you to do a followup study, you could survey a random sample of (say) 100 incoming students at your (or some other) school and ask them to answer a question or two.

you could then do a repeat survey 3 or 6 monthes later with a different random sample of 100. (you could also contact the same 100 as before - if you can get hold of them.) then you could see if attitudes or opinions expressed in the first survey have changed.

the method of analysis would be somewhat different if you use the same students in the repeat survey than if you take a new sample. in either case, standard analyses (for independent or paired-sample data), as given in most statistics texts, can be used to analyze the data.

Answer 2

How about this one?

Collect the closing Dow Jones Industrial Average (DJIA) for some period, say, sixty market days.

Collect the closing futures price per barrel (p) on crude oil each day over that same period.

Test the following null hypothesis at the 95% confidence level:

H0: The DJIA is not related to p; that is, rho=0

If the DJIA is not related to p or if it is related to p (at 95% confidence), explain why...what is the cause effect if any?

PS: The next answerer is right about "be careful." What he/she is referring to is called "post hoc fallacy." These words simply mean that just because you can show correlation statisticially, that doesn't mean there is a real cause-effect. This is why I wrote "what is the cause effect if any" There may be none that you can explain by physical cause and effect.

I like to use the example of the crowing rooster. You can show very high correlation (rho = 1.0) between the rooster crowing each morning and the sun coming up. Did the rooster CAUSE the EFFECT of the sun rising? Clearly not, if anything, the sun caused the rooster to crow; not the other way around.

So, bottom line, if you do the suggested experiment and find high correlation, the next step is to review how the oil futures might have caused the DJIA to vary with the change in oil prices. For that you will have to know something about supply and demand, and how the market prices and index are set.

Answer 3

I think that it is related to regression. effect of X on Y. For e.g effect of price on demand. Y=a+bx

Answer 4

i think by using regession analysis or blocking design such rcbd, crd, latin square. you may use it by related to your study case.. or you may use transformation mathods if the data is not significant.... just try... may be .. who knows.. right..