@ Larry Mintz, If sample size is small one can use small sample tests like chi-square test, t test, F test, AOV technique etc. , in addition to non-parametric tests, depending upon the type of study.

However for large sample (i.e. for sample size more than 30), z test i.e. normality test can be applied.

The biggest detriment to normalizing data os the artifically increased SEs of your rextreme values. If you care abiut your extreme values or have a pltykurtic distrivution, normalizing will essentially reduce the impact of these extrene values.

Matt

You have not given any graphs or other information suggesting that normalization may of use. Not enough information. David Booth

This is a single point data. Nothing can be done on a single data.

Hi Masoud,

In your description the combination of the "time" and "street" variables suggests that the design may involve repeated measures for each observation--that is, for each street(s). In repeated measures designs, ipsative normalization is useful in studies involving multiple observations (i.e., two or more streets):

https://odajournal.com/2013/10/23/ipsative-standardization-is-essential-in-the-analysis-of-serial-data

https://odajournal.com/2016/05/31/using-machine-learning-to-model-dose-response-relationships-via-oda-eliminating-response-variable-baseline-variation-by-ipsative-standardization

Ipsative normalization is also useful in repeated measures studies involving a single observation (i.e., only one street):

https://odajournal.com/2013/11/07/ascertaining-an-individual-patients-symptom-dominance-hierarchy-analysis-of-raw-longitudinal-data-induces-simpsons-paradox

I recommend normalizing mainly variables related to physical magnitudes, which, by their nature, should show a normalized behavior and therefore the problems derived from their normalization would be minimal.

As for your question, I would say "not necessarily". In regression analysis, we generally normalize the data if heterogeneity occurs. Rest, the information by you is not enough to conclude in particular.

It is not necessarily. It depend on the structure of the study.

It is not necessary to normalize a given data set always.

However, sometimes it becomes necessary.

It depend upon many factors like type of data, objective(s) of study, method(s) to be used in analysis etc.

If it is small use non-parametric methods

As n gets large n tends towards the Normal

distribution, so then Normalization is good

idea to smooth the errors.

Remember the Central Limit Theorem and

Law of Large numbers

ex.

Dataset size n=10, assume Poisson dist

E[X]=Var[X]=5

By CLT.

Suppose we want to estimate

we have 10 iid variables and we want to

what might happen if we get 15 more

variables,Normalization a good idea

Pr{X1+...+X10>15}

By CLT

P{X1+...+X10-5)/10*sqrt(5)>P{(15-5)/22.36}

~P(N(0,1)>1)

~1-0(.447)

= 1-.0.67

~.33

Adding more variables suggest normalisation would be a bad idea

since p

@ Larry Mintz, If sample size is small one can use small sample tests like chi-square test, t test, F test, AOV technique etc. , in addition to non-parametric tests, depending upon the type of study.

However for large sample (i.e. for sample size more than 30), z test i.e. normality test can be applied.

You might use any GLzM or equivalent that accepts several distribution types as well as is you gain in robustness

@ Masoud Masoumi Moghadam ,

Specify/define the dependent variable and the independent variable of your study. Then collect data on these two variables. Only then it is possible to search for a suitable model for your prediction purpose.

I don't think that normalizing is necessary! You can only normalize when you notice that either one or both of parametric assumption(s) is/are violated. The information you've provided in your question are not enough; you need to revisit the concept of regression again, find out if the assumption(s) stated above is/are violated, then to avoid drawing a narrow conclusion, you can normalize that data set.

I hope this will help.