Find Trend in STAT: How Many Observations Needed?

[knightza]

Hey, does anyone know how many observation is needed before we can prove that a trend is significance?

 

[Cyrus]

You don't prove a trend is significant based on the number of observed points, its based on the confidence level.

 

[knightza]

Hey Cyrus, thanks for your reply, sorry I'm not very good at STAT here but I currently have to analyse quite a huge set of data. The first lot they sent me has around 7000 observations (and I'm trying to relate the temperature with speed, pressure etc) so how many "strange observations" am I allowed if, say, i want it to be with in 95% confidence level. Is there some formula to calculate this??

Thanks

Knight

 

[Cyrus]

What you are trying to do is system identification. There is a large body of theory behind it, along with statistical metrics to determine what terms to include in your model for explanatory variables. If you do not know how to do this, I would recommend you give it to someone that does because this is not something you can teach yourself in short order and it requires someone trained in SYSID.

It doesn't work based on "strange observations," I think the problem is that you don't have a fundamental background in statistics.

 

[knightza]

OMG this is so..sad. why did they give this research project to a third year uni student T.T

Ok, thanks for your help anyways, aye? I might need to talk to my supervisor about it.

 

[Cyrus]

OMG this is so..sad. why did they give this research project to a third year uni student T.T

Ok, thanks for your help anyways, aye? I might need to talk to my supervisor about it.

It all depends on what your data looks like. If the data has very clear distinct trends, then you can probably get away with using some simple functions and seeing what the R^2 value looks like. If you have a more complicated system, then things will be nasty pretty quickly. Do you know what the time history of your signals look like?

 

[knightza]

What did you mean by time history though?

This is quite a specific project on a specific mine site and the data was just recorded every minute since 25th November.

I am now waiting for the second lot of data since it was found that the data on the first lot contains many errors in it (due to configurations, installation of the device and so on).

Hopefully the relationship will not be that complicated. I'll certainly get back to you once I have more problems (most likely when I get the second lot of data).

Thanks

Knight


PS

One thing that I need the clarification for, say, if we are looking at the data from the tyres of one vehicle. Should I be analysing the data of each tyre separately? both temperature, pressure etc. Or is there a possibility that the increase in pressure/temperature of one tyre will cause an increase in the other tyre too?