Normality test temperature data

Hi,

I have a column of temp data - some neagative values.

I applied a KS and lillietest to the data - it is NOT normally distributed.

As data has negative values, I can't apply a log transform.

Even if I apply log transform, the data takes on non-real values e.g. 5.21493575760899 + 3.14159265358979i

Any ideas how I can transform the data to make it normally distributed?

Nothing too complicated - I am a beginner.

Thank you.

[dpb]

"Any ideas how I can transform the data to make it normally distributed?"

W/O any idea of what the data looks like, no.

But, there may be no need...what is end objective?

However, if you feel compelled the Box-Cox transformation is one common technique that is pretty general. See <NIST handbook EDA Box-Cox Transformation>

Answer 1

[Ridwan Alam]

Since linear shift won't affect the distribution, you can shift the data and perform log transformation.

%%  test data
data = randn(1000,1);
data = data - min(data) + 1;
% making sure that the data is not normal 
data = exp(data);
% also, since your data has many negative values
data = data - 0.9*max(data);
lillietest(data) % ans = 1

% LOG TRANSFORMATION

% first, shift your data to all positive range
posdata = data - min(data) + 1;
lillietest(posdata) % ans = 1
% then, perform log transformation
logdata = log(posdata);
lillietest(logdata) % ans = 0

Hope this helps!