Skip to content

lm and time series.

8 messages · rkevinburton at charter.net, Gabor Grothendieck, stephen sefick +1 more

rkevinburton at charter.net
# 
I did a ?lm and it said basically to be careful when using lm and a time series. But my question is probably more to do with my inexperience that anything. If I have a time series object 'ti' how do I write the formula? The response is the value at any particular time and the time is basically the index of the time series. But I don't know how to put that into a formula.

Thank you.

Kevin
Gabor Grothendieck
# 
The Time Series section in ?lm should be self explanatory.   If you are using
diff's and lag's then look at the dyn package.
On Fri, Sep 5, 2008 at 12:25 PM, <rkevinburton at charter.net> wrote:
rkevinburton at charter.net
# 
I am sorry but I looked at ?lm and could not see any guidance on writting a formula. If I have two arrays or a data set then I know how to do that (y ~ x) but for a time series I am not sure how to write y or x.

Thank you.

Kevin
---- Gabor Grothendieck <ggrothendieck at gmail.com> wrote:
stephen sefick
# 
what do you want to do?
On Fri, Sep 5, 2008 at 3:22 PM, <rkevinburton at charter.net> wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      rkevinburton at charter.net
    

    

      
      #
    

  


  

      
I want to fit a function to time series. If I had:

x <- 1:4
y <- 1:4

lm(y~x)

This would fit a simple line to the four points. But if it is represented as a time series

x <- 1:4
t <- ts(x)

lm(????)

So I have a time series in the object t. How do I write a formula for lm? What do I put in the formula for x and y when I only have t (the time series). 

Kevin

        
---- stephen sefick <ssefick at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      stephen sefick
    

    

      
      #
    

  


  

      
So you want time as the independent variable?  Let's say that the
units of y in your first example were seconds- couldn't you just use a
regular lm and say that the units were seconds, minutes, or what ever?
 I am probably out of my league here, but I am just not understanding
what it is that you want.  a time series is just a series of data
points indexed by time.  Arima maybe, or some other cool times series
modeling approach- wavelet, spectral density- for frequency domain
type things...  What are you trying to accomplish?

        
On Fri, Sep 5, 2008 at 5:47 PM,  <rkevinburton at charter.net> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      rkevinburton at charter.net
    

    

      
      #
    

  


  

      
That is the thing. As a new comer to 'R' I don't understand how to write a formula when all I have is a time series. I don't know how to express the independent and dependent variables in a formula when the object is a time series. So please just solve this simple example and I will extrapolate from there.

Say the units of the time series is days and the value at each point is the response. If I wanted to fit a straiight line through the following time series:

y <- 4:7
t <- ts(y)

So this is saying to me something like 4 units were sold on the first day, 5 on the second, 6 on the third, and 7 on the fourth.

So given the time series t I want to find the slope and inercept:

y = m*x + b

with x in days and the respoinse would be the number of units sold. I need to find 'm' and 'b'. If all I have is t (the time series above) then what would be the formula, and for that matter the arguments to lm to give the desired result?

fit <- lm(???)

Thank you.

Kevin

        
---- stephen sefick <ssefick at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Achim Zeileis
    

    

      
      #
    

  


  

      
On Fri, 5 Sep 2008, rkevinburton at charter.net wrote:

        

            

      
      
      
    

        
y <- ts(4:7)
lm(y ~ time(y))

And as previously pointed out to you: To preserve the time series 
properties, look at ?lm, the "dynlm" and "dyn" packages.
Z