I am always confused when I am making null and alternative hypothesis,
I know we intend to formulate null hypothesis with the intention of rejecting it.
But when ever I have a statement i get confused that which is which.
Is there any hard and fast rule that I should keep in mind.
Like in the statements below, I just believed my whims, can you see if I have got any sense of how to make the null and alternative?
It is just practice not home work so you can take your time to reply me.
1.In a highly publicised murder trial it was thought that 25% of all TV viewers watched the proceedings on TV. An analyst felt that this estimate was too small. He selected a random sample of 100 and found 32 of them actually watched the proceedings. The appropriate alternative hypothesis would be
H0 : p=0.25
H1 : p>0.25
2.A member of the Islip Urban Renewal Taskforce claimed that the installation of internal heating facilities in homes had led to the proportion of homes with fireplaces falling below one half.
State the direction of the alternative hypothesis used to test the taskforce claim.
H0: p= 0.5
H1: p not equal to 0.5
Another committee member, who did not originally believe the claim, took a random sample of 84 houses in the area to test the validity of the claim. He found that 33 of them had fireplaces.
H1 p< 0.5
SO do I have any sense of how to make null and alternative?
Last edited by Dason; 10-18-2010 at 12:15 PM.
I remember being confused by this null hypothesis thing when I first learned statistics, so don't feel alone.
I can see how one might think that the null hypothesis is always "the opposite of what you want to prove". Lots of the standard examples that one gets in introductory statistics classes are indeed like that. But that isn't the rule. The real rule is that the null hypothesis has to be specific enough to predict the distribution of your test statistic.
For example, "the drug has an effect" isn't a good null hypothesis for a 2X2 experiment because there are a zillion different ways that the drug could have an effect, each of which would predict a different distribution of counts. But "the drug has no effect" predicts a very specific distribution of counts. So in that case your null hypotheis does indeed turn out to be "the opposite of what you want to prove".
Consider, though, the case of testing the goodness-of-fit of a model to experimental data. In this case, "the model doesn't explain the data" isn't a good null hypothesis, because there are a zillion different alternative models, each of which would predict a different distribution of measurements. So in this case the null hypotheis is "the model explains the data", which, again, makes a specific prediction about the distribution of measurements. Notice that, in this case, the null hypothsis is "what you want to prove".
OK, got it , a little but still , So I am thinking null something that we have to prove regardless of claim, like if someone claims I have iq=136, and I have to prove no he has less than that ..
so my alternate will be iq< 136...but in this case less than is my alternative not null
Advertise on Talk Stats