[R] test hypothesis in R

[David Winsemius]

> On Mar 23, 2016, at 1:44 PM, ruipbarradas at sapo.pt wrote:
> 
> Hello,
> 
> Try
> 
> ?t.test
> t.test(mA, mB, alternative = "greater")
> 
> Hope this helps,
> 
> Rui Barradas
>  
> 
> Citando Eliza Botto <eliza_botto at outlook.com>:
> 
>> Dear All,
>> I want to test a hypothesis in R by using student' t-test (P-values).
>> The hypothesis is that model A produces lesser error than model B at  
>> ten stations. Obviously, Null Hypothesis (H0) is that the error  
>> produces by model A is not lower than model B.

NOT "obviously". You only get to do one-sided tests when the scientific question would not allow the possibility of a departure to "the other side".

Two-sided tests are the norm in scientific literature, often to the experimenter's distress when they haven't done a thoughtful (non-optimistic) power analysis and their results are inconclusive as a result. Your hypothesis _should_ have been constructed _before_ you saw the data. That is if you want to be an ethical scientist.


>> The error magnitudes are
>> 
>> #model A
>>> dput(mA)
>> 
>> c(36.1956086452583, 34.9996207622861, 36.435733025221,  
>> 37.2003157636202, 36.1318687775115, 37.164132533536,  
>> 35.2028759357069, 36.7719835944373, 38.3861425339751,  
>> 37.4174132119744)
>> #model B
>>> dput(mB)
>> 
>> c(39.7655211768704, 40.1730916643841, 39.3699055738618,  
>> 39.401619831763, 41.1218634441457, 39.1968630742826,  
>> 40.5265825061639, 40.4674956975404, 40.5954427072364,  
>> 41.4875529130543)

Those are not models. They are just vectors of numbers. And they seem unlikely to be residual errors of a linear model since they are not centered on zero. I doubt there is enough in your presentation for a sensible comment on the proper analysis.

-- 

David.

>> 
>> Now can I test my hypothesis in R?
>> Thankyou very much in Advance,
>> Eliza
>>         [[alternative HTML version deleted]]
>> 
>> ______________________________________________



David Winsemius
Alameda, CA, USA