1. ## A bit of help..please?

I am not sure if anyone here can help me at all, or if this message is in the correct place, but here i am anyway.
I use Harvard Graphics AP to produce vertical bar graphs - I show the months results so far and a target point. I produce a trend line to show where we are heading - i just use another column with the months results in again, and turn these into a "line" with a "trend" fit. Dead easy no one has questioned it...until now!!!
Basically, this trend line does not calculate a trend like I, or anyone else here would! I would take the sum of the 4 months so far, divide by 4 (being April's figures) and the multiply by 12 to get an end point for the year. I do this and i get to a different answer to Harvard!!
I am wondering if somehow it takes into acconut that the monthly figures are falling / rising and so adds / subtracts a bit?? Does anyone know how Harvard might be calculating this trend line???
Thanks for any help,
Al

2. ## Re: A bit of help..please?

I don't have Harvard Graphics, but the way a trend line is usually calculated is by the so-called "least squares" method. If you draw a straight line, you can measure the vertical distance from the series points (I have drawn the points as big dots in the attached picture) to the line. These distances can be positive or negative. The square of the distances is always greater than or equal to zero. Using a bit of high school mathematics, it is possible to calculate for which line the sum of these squares is minimal, i.e. which line best "hugs" the points. This is the trend line.

3. ## Re: A bit of help..please?

I never knew that you were supposed to measure vertical distances, I always thought you measured the distance from each point to the line along a line perpendicular to the trend line.

StuartR

4. ## Re: A bit of help..please?

Of course, variations are possible - perpendicular distances would be one, absolute values instead of the square of the distance another. The "standard" definition uses vertical distance because that indicates the error in the predicted value. Using the squares of these distances makes it fit in nicely with statistical measures such as standard deviations and correlation coefficients. The slope of the trend line corresponds directly with the correlation coefficient of the x and y values.

5. ## Re: A bit of help..please?

Hmmm....thanks for your help fellas! Unfortunatly there is never a high school kid around when you need one is there?!?! This does ring some dim and distant bells in my head - I soon as i left school though i completely forgot everything i'd ever learnt! (How was I to know it'd be useful one day?)
Thanks again - I'm gonna have a play with some numbers and graphs now.
Al. <img src=/S/dizzy.gif border=0 alt=dizzy width=15 height=15>

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
•