Please see that I have reversed the x axis, so as just to make the plot clear. Or else the graph looks like as attached below (the x axis is as in my first 3D ball-graph). Yes the matrix had 100 values. Regarding spacing please see another ball graph attached below.

You are right. sure. I will reverse the color scale..

Hello, have you tried matlab?

x=[25 25 25 50 50 50 50 50 50];

y=[30 40 50 30 40 50 30 40 50];

z=[1.817 1.763 1.716 1.748 1.659 1.648 1.654 1.563 1.561];

xx=[x(1:3)',x(4:6)',x(7:9)'];

yy=[y(1:3)',y(4:6)',y(7:9)'];

zz=[z(1:3)',z(4:6)',z(7:9)'];

surf(xx,yy,zz);

I have a little question. Why the coordinates of X and Y are the same in the following 6 lines in your data but the Z coordinates different?

Sorry, I forget the figure.

There are too few data to make a reasonable surface plot. I would find a heatmap more appropriate. Or even a simle line plot (with 3 lines either Z vs. X for each Y or Z vs. Y for each X) And I also wonder if the duplicate x/y coordinates are intended or a mistake.

Hi Yu and Wilhelm,

Thanks for your reply and information. Please note that the X and Y coordinates are not same!! Please look at them carefully, the values are different and changing periodically. Actually, I am looking to have a surface graph something like which is attached below..???

You have identical coordinates. For example:

Line 4: 50-- 30 --1.748

Line 7: 50 --30 --1.654

And the plot shown seems to be the result of a fitted model, quadratic on ethanol and linear on temperature, including an interaction. If wou have such a model, you can calculate predicted response values of any combination of the predictors. If you do this for a reasonably dense grid of values (the plot shows a 10x10 grid) you can use any spreadsheed or math software to plot such a surface. Some math/stats software will plot you the surface directly when you just provide the model formula (e.g. Mathcad or Mathematica and numerous specialized 3D function plotters).

Edit PS:

If you do not have a model you can get such a kind of a smooth surface using spline interpolation (spline surface). But I am still of the opinion that you have too little data - such a 3D surface plot would not really add insight, it might rather highlight structures that are easily over-interpreted, especially if you do not have a model (that should be justified *outside* of the data, by logical, scientific reasons [not data-driven]). A more or less arbitrary 2D smoothing by a surface is no better than some weighted average with more or less arbitrary weighting factors in the one-dimensional case.

Thanks Wilhelm.. You are right. The attached plot they have obtained using a fitted model. I too have obtained following data fitted to mathematical equation. Could yiu please tell me how to obtain 'predicted response values'?? give me little more details. Mean time I plotted the above values in Origin software using XYZ 3D symbol graph. Do you think it gives sufficient information on variation of conc. with respect to temp. snd solvent composition?

The predicted response is simply the value of the fitted function (at the desired combination of predictor values). To keep it simple, consider a simple linear regression. The model is Y = aX+b. The values of the coefficients a and b (slope and intercept) are determined by fitting the model to the data. Say, for instance, these values are a=2.3 and b=152. The you get the predicted value for X=10 by simply plugging this value into the function: Y = 2.3*10+152 =175. You can so calculate the predition for any value of X you wish. A 2D-model is a function like Z = aX²+bX+cY+dX²Y+eXY+f (taken from the example in your previous post). Again the coefs a,b,c,d,e,f are obtained from fitting the model. The you can put in any combination of X and Y and so calculate the predition (Z) at this point.

Your lates plot shows that there is not much variability in the "temp." values, and that this littel difference in "temp." has not much influence on the response. Actually, your whole data is best to be described only considering "solvent compo." (and ignoring "temp.", since it does not add information here). But the it reduces to a simple linear regression (it even seems that a straight-line fit might be sufficient, what would be a model like "conc". = a*"solvent compo." + b. If you have a curved relation you might consider a parabola ("conc". = a*"solvent compo."²+ + b*"solvent compo."+c) or -best solution!- you have a physical or biochemical principle in mind that would suggest a specific functional form of this relationship.

I don't recommend Excel. Matlab is the best option, have you tried to do it in the muPad app in Matlab? It's very user-friendly to set up different viewing options, rotating and layout options as well.

Regards

@ Dr. Wilhelm, you have given me some good ideas. I tried the technique with Origin Software. I converted the data into matrix of 10X10. Then I got a nice graph with a surface response as shown below. I hope this is good. What do you say?

Well... looks pretty... but there is nothing more I can say. I just notice that the plot does not fit to your previous plot (showing 10 points) - already considering that the directions of the axes have changes. I further wonder if the matrix converion was ok. Do you have 100 values? and are they equally spaced in each dimension (solvent compo and temp)? Finally, I would revers the color scale at the right side (lowest conc value at the bottom, highest at the top of the scale).

Please see that I have reversed the x axis, so as just to make the plot clear. Or else the graph looks like as attached below (the x axis is as in my first 3D ball-graph). Yes the matrix had 100 values. Regarding spacing please see another ball graph attached below.

You are right. sure. I will reverse the color scale..

Equally space matrix values

Your latest surface seems to fit to the 10 values you showed. Seems corred now. Have you *measured* 100 values, at these 100 combinations of solvent compo and temp? If so, everything seems ok. But I doubt that this is the case. Can you also plot all the values you have actually measured in the surface plot?

No, I have not measured all the 100 points practically!!!! I have measured only those combinations mentioned in the beginning. Do you think it is important to measure all the 100 combination practically???

No need to have all 100 values - these can be interpolaions. But where is the model from? Did you fit it to these 10 points shown above? If so, then I think that the model is simply pretending to provide more information than you have - and I am talking about the curvature. But since it is relatively "flat" (staight), this should not be a major concern. I would show the values used for the fit in this surface plot, just to visualize on what actual data the model is based upon.

Yes, you are right. Now i think it is better to highlight the point in the surface plot which are actually used in the experiment. Right?

surf(X,Y,Z)

the above function can solve your problem.

OR follow the below link

https://www.mathworks.com/help/matlab/ref/surf.html