What statistical test to use to examine the relationship “Earnings” vs “Gender”

[lolol0101]

Newbie at SAS, I did as

proc freq data=want;
tables earnings*gender / chisq;
run; 

but I doubt it's right because I went to google the methods

to analyze its results?

Thank you.

 

 

 

[MarkusWeick]

Hi @lolol0101,

my advise would be not to start with test but with (graphical) analysis of the data. For example the frequency distribution of earnings, one per gender. Comparing the distributions might give some ideas about a hypothesis that than can be tested.

So, do you want to show us graphs of the frequency distributions?

Best

Markus

[Ksharp]

TTest
PROC TTEST
or
Wilcoxon test (nonparameter version of TTest )
PROC NPAR1WAY .... wilcoxon ;


Also calling @Rick_SAS @StatDave to get more statistical method.

[lolol0101]

The gender has 3 levels so I would get the error "class variable has more than two levels"

is T-Test still appropriate to use?

[Ksharp]

Then try ANOVA .

proc glm data=have;
class gender;
model salary=gender /solution;

lsmean gender/diff pdiff adjust=dutt;
quit;

The better way is non-parameter version of ANOVA as Paige said.
or you could try GAMMA distribution since your salary is all positive.
proc genmod data=have;
class gender;
model salary=gender/dist=gamma;
lsmean gender/diff pdiff adjust=dutt;
run;

[lolol0101]

Thank you! last question, what if it's "salary" vs. "age"? (both are numeric)

How does one examine this?

[PaigeMiller]

---

@lolol0101 wrote:

Thank you! last question, what if it's "salary" vs. "age"? (both are numeric)

How does one examine this?

---

1. Scatterplot
2. Regression

[Ksharp]

As Paige said , Regression .(if you have big data ,also could try ANOVA)
And I would recommend to use violin graph .

https://blogs.sas.com/content/graphicallyspeaking/2012/10/30/violin-plots/

[PaigeMiller]

---

@lolol0101 wrote:

Newbie at SAS, I did as

proc freq data=want;
tables earnings*gender / chisq;
run; 

but I doubt it's right because I went to google the methods

to analyze its results?

---

I would say use the Kruskal-Wallis test in PROC NPAR1WAY, which uses the Wilcoxon scores. Example.

This is the non-parametric equivalent of a one-way ANOVA. A chi-squared test is not really appropriate for the (almost) continuous variable earnings.