Edit the limits in least squares line.

Hi everyone!

Plotting the lsline in a scatter plot I get the least squares line but it is stretched out of the period with the data as can be seen below. Is there anyway I can have it only above the period of interest (the period that I have data, 1976-2016)?

Thanks in advance!

[dpb]

Sure, just evaluate the fit at the first/last points.

Problem with a specific solution is we don't have a klew how you generated the above plot and there are any number of ways in which could have done...show precise code that created the above.

[Dimitrios Bentis]

I am not sure how to evaluate the fit at the 1st and last points. The code that generated the graph above is this:

figure(4)
plot(dt,WD_autumn,'o')
lsline
ylim([0 360])
datetick('x','yyyy','keeplimits')
xlabel('Year') 
ylabel('Wind Speed m/s') 
title('autumn')

[dpb]

Well, unfortunately, lsline doesn't help much for anything other than the default case; there's no way to find out what the coefficients are so nothing you can do with it.

That kind of thing is why asked about how you'd created the line...

[Adam Danz]

Using polyfit or fitlm is the way to go but because I'm drawn to finding alternatives, here's a workaround that only relies on lsline.

Before calling lsline, set axis tight or set xlim to the range of your data, then call lsline, then set your desired axis limits.

x = linspace(1975,2017,50);
y = rand(size(x))*50+130; 
plot(x, y, 'o')
axis tight
lsline
xlim([1970,2020])
ylim([0,350])

Answer 1

[dpb]

If you need to do more than what lsline does which is to draw the line from the axis limits but not give anything else useful for modifying it; then you have to fit the line itself. I really don't understand TMW's thinking on a bunch of this stuff... :(

You've got a time axis; none of the fitting routines I tried here with R2016b are datetime aware so will illustrate with datenum; same idea if you have a more recent release and does work with datetime

>> dn=datenum(1975,1:3:128,1).';        % a stretch of datenums 
>> w=50*rand(size(dn))+140;             % some data to go with it
>> plot(dn,w,'0')
>> ylim([0 350])
>> datetick('x','keeplimits')       
>> lsline             % put the default lsline on for comparison later...
>> b=fitlm(dn,w,'linear')  % do a linear fit...
b = 
Linear regression model:
  y ~ 1 + x1
Estimated Coefficients:
                 Estimate        SE         tStat      pValue 
                 _________    _________    ________    _______
  (Intercept)      -1333.7       1376.8    -0.96873    0.33836
  x1             0.0020712    0.0019036      1.0881    0.28291
Number of observations: 43, Error degrees of freedom: 41
Root Mean Squared Error: 14.1
R-squared: 0.0281,  Adjusted R-Squared 0.00436
F-statistic vs. constant model: 1.18, p-value = 0.283
>> hold on  % to add to the plot
>> hLS=plot([dn(1);dn(end)],b.predict([dn(1);dn(end)]),'r-');  % add the LS fit over data range

This gives:

so you can see is the same line just limited to the range of the data.

Answer 2

[Scott MacKenzie]

As noted, lsline isn't very useful beyond its default behaviour. To constrain the limits of the regression line, a simple option is to use polyfit and line, as below.

x = [5 7 3 8 6 9];
y = [4 5 3 6 5 7];

scatter(x,y);
axis([0 10 0 10]);

p = polyfit(x, y, 1);
m = p(1); % slope
b = p(2); % intercept
line([min(x) max(x)], [m*min(x)+b m*max(x)+b], 'color', 'r');

This yields