Level of significance

[Emeka Dim]

Please, I was conducting a multinomial logistic regression analysis and I saw several variables with substantial odds ratio with p-values at 0.053 and 0.056:
Variables Odds ratio Std Error z P>|z| Confidence Interval
Severity1 | 2.375857 1.077781 1.91 0.056 .9765223 5.780407
Emo_impact | 1.206141 .116789 1.94 0.053 .9976482 1.458207

`Severity1' was measured as a dichotomous variable while `Emo_impact' was measured at the interval-ratio level. The dependent variable has three categories of which the first category (no occurrence of incidence) was used as the base category for the multinomial logistic regression. The figures above represent the output for one of the categories of the regression output

There are five other variables in this multivariate logistic regression which are not shown and the number of observation is 380. The Pseudo R2 is 0.2214, Wald chi2 (18) = 83.08 (Prob > chi2 = 0.0000).

I know that the rule of thumb for identifying significance levels is 0.05 but I felt that these variables have substantial odd ratios and p values. Also, the n-size may have obscured the nature of the relationship between the two variables.

Please, would you advice one to consider any or both of these variables as significantly associated with the dependent variable?
I can provide more information for the table if required.

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

You are mixing two very different things. One is statistical significance - given the estimate of the parameter, can we reject the hypothesis that the true value is zero (a function of the size of the parameter and how accurately the parameter is estimated). [Note - this is my casual statement, I suspect some would offer a more accurate description of statistical significance.] The other is practical significance - how much the predicted value changes for a change in the iv or how much of the variance in the dv the iv explains. Note that all the parameters depend on the scaling of the variables - you could divide severity by 100 and drastically change the odds ratio. So, you have to be aware of what a one unit change means in the different variables. Some look at the change in the dv for a one standard deviation change in the iv.

There is a substantial community that objects to the p-values. Why is .050 good but .051 bad? Why the arbitrary .05 or .01 levels? Why test against zero rather than some alternative reasonable value? I'd report of full p value and then interpret the parameters (noting the issue that it just misses the .05 hurdle).