I have continuous data that are left skewed and on a negative range [-10,-2]. Would it be a valid approach to just mirror the data by multiplying by -1 and then try to fit distributions on the positive range? Eventually, I want to sample from the fitted distribution so would it be ok to just multiply a sampled value by -1 again to get it on the negative range?
$\begingroup$ Why is it easier to model on the positive range? $\endgroup$– DaveApr 11, 2020 at 18:08
$\begingroup$ If you flip the data they become right skewed and they are in the positive axis, so I can use any of the common distributions e.g. lognormal, gamma, exponential to try to fit. $\endgroup$– nicnazApr 11, 2020 at 21:03
$\begingroup$ What does the data represent? $\endgroup$– Georg M. GoergApr 11, 2020 at 21:56
$\begingroup$ It is household expenses $\endgroup$– nicnazApr 11, 2020 at 22:01
If your data is positive, you should state this in your question. In that case, you can use any flexible distribution with support on the positive real line, or to transform the data using the logarithm and use a flexible distribution distribution with support on the entire real line.
There are many distributions with support on the positive real line that can capture positive and negative skewness. See for instance the review paper:
$\begingroup$ My data is negative as I stated it is in the range [-10,-2]. By flipping them I can use any of the gamma, lognormal etc distributions to fit them. But I am not sure if this is a valid approach? $\endgroup$– nicnazApr 11, 2020 at 21:02