Trendline

Can you upload the data you're using instead of just the chart image?

Your question makes sense now, but the intercept you are setting is the x-intercept, not the y-intercept.

I think there's a few things to consider before taking this route and using the results for your analysis:

1) Excel does not natively allow chart trendlines to go through the origin because that inherently takes away the point of a trendline - to match the data given. If your trendline is behaving poorly, this is usually a result of poor data (or rather, poor instrumentation). It is important to note that by altering a trendline to go through the origin fundamentally alters the forecasting capabilities of the trendline, whereby you degrade what the capability is actually used for.

2) It is not uncommon when testing empirical data that you void the first few samples of data while the test runs. This is to reduce noise and obtain precise results. By inspection, it appears this is what the case is with your data, so I would recommend starting with row 4 data in your sample workbook instead of row 2.

3) Ultimately, since this is your data and not mine, I can only give recommendations; after all as far as you know, I'm just some schlub on the internet. If you are truly set on creating a trendline through the origin, I know LINEST can be used, albeit through some manipulation. This link describes a similar process I have used: http://www.oooforum.org/forum/viewto...ight=trendline

Hope this helps!

The R² value shows you how close to the trendline value your data is. The closer to 1.00, the more accurate your data is to the linear trendline fit. However, this is your data. I don't know how it was collected, manipulated, or what it even pertains to. That said, I can only comment on how the data should look, given that you said it should have a linear relationship.

Because your data ranges are different sizes (based on your sample workbook), I'd be cautious in using R² values. Upon inspection, it looks as though Node.js CPU Average should follow the linear trend better than Eriang CPU Average. However, when starting the data values at 40 target users / second, the R² values respectively are 0.9714 and 0.9989.

Even though the margin of difference between the two is ~0.02, it's still counter intuitive to think that Eriang CPU has a more linear trend than Node.js CPU...until you realize that the data ranges are different, and the population size of Node.js is bigger than Eriang, meaning more possibility for deviations in data.

I know it's a mouthful, but what I'm trying to say is statistics can be skewed all different kinds of ways to get the results people are after. This is a main reason the scientific method is so stringent on its statistical analysis policies; to avoid potential bias and misinformation.

Here's the real problem that I see (and it is one of the most common mistakes I see in spreadsheet charting) -- you are using a line chart (which means a category x axis) rather than a scatter/XY chart. Excel is not seeing your x values as number, but it is seeing them as "category1", "category2", "category3" and so on, and it is using those count numbers (1,2,3,...) as the x values in the trendline regression. Thus x=0 is one category to the left of your chart and not the actual 0 value at all. I think this problem resolves itself quickly and easily if you will change your chart type from line to XY scatter, so that Excel will see your x values as actual numbers and use them in the regression.

You might review this for a discussion of the differences between line and XY scatter chart types: http://peltiertech.com/Excel/ChartsH...ValueAxis.html

Originally Posted by MrShorty

Here's the real problem that I see (and it is one of the most common mistakes I see in spreadsheet charting) -- you are using a line chart (which means a category x axis) rather than a scatter/XY chart. [/url]

Note to self: Check chart type first. Save the headache.

That looks to be correct.