1
$\begingroup$

Descriptive Statistics

I'm trying to make an assessment of these figures, with the aim of using them for multivariate regression. I'm struggling and so far all I have is:

"With the lowest skew, kurtosis closest to zero and a mean most close to its median, average return is the closest to a normal distribution. PTBV with a mean so far from its maximum is likely to contain outliers"

Is what I've written correct? What are the questions I need to be asking myself to get the best interpretation of this data?

I've searched online/on here for some analysis of descriptive statistics but have struggled to find anything 'juicy'.

I'd be ever so grateful for some ideas.

$\endgroup$
5
  • $\begingroup$ Have you thought about trying to represent some of these features graphically? Boxplots, histograms, scatter plots all reveal a great deal. Also, for the variables you wish to use as explanatory in your regression be sure to check the correlation matrix. Multicollinearity is always lurking in the background. $\endgroup$ – JohnK Jan 8 '15 at 1:08
  • $\begingroup$ Hi John, thanks for your comment. I have made scatter plots, histograms, a correlation matrix and actually finished the multiple regression model (OLS). It's just my assignment requires a discussion on these Descriptive Stats (I was supposed to do it before the final regression) and I'm struggling to find something to write about. I feel like I don't know what to look for. $\endgroup$ – Harry Jan 8 '15 at 1:17
  • $\begingroup$ You can comment on the location, e.g. mean median and the spread of these variables. Then see if there are outliers, using the boxplots. In the correlation matrix of the predictors, what are the highest correlations in absolute value? Just some ideas. $\endgroup$ – JohnK Jan 8 '15 at 1:20
  • $\begingroup$ Just an observation that data representation with too many decimal places (unless it's necessary) hurts readability. If you're not scared of reading R and LeTeX code, this nice example in RSweave IMHO illustrates not only how to do EDA (and regression) write-up, but how to do it in a reproducible research fashion. $\endgroup$ – Aleksandr Blekh Jan 8 '15 at 4:18
  • $\begingroup$ Useful discussion of the descriptive statistics is likely to require reference to what the numbers represent: Are negative values of PTBV qualitatively different from positive values, or even possible? Will the model need to extrapolate beyond the limited range of VaR values present in the sample? Could the low average return indicate a selection bias? &c. $\endgroup$ – Scortchi - Reinstate Monica Jan 8 '15 at 14:17
1
$\begingroup$

Partially answered in comments:

Have you thought about trying to represent some of these features graphically? Boxplots, histograms, scatter plots all reveal a great deal. Also, for the variables you wish to use as explanatory in your regression be sure to check the correlation matrix. Multicollinearity is always lurking in the background. – JohnK

But note that it is often better to start with a simple regression model, and then plot residuals from that model.

You can comment on the location, e.g. mean median and the spread of these variables. Then see if there are outliers, using the boxplots. In the correlation matrix of the predictors, what are the highest correlations in absolute value? Just some ideas. – JohnK

Useful discussion of the descriptive statistics is likely to require reference to what the numbers represent: Are negative values of PTBV qualitatively different from positive values, or even possible? Will the model need to extrapolate beyond the limited range of VaR values present in the sample? Could the low average return indicate a selection bias? &c. – Scortchi

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.