Not Significant Coefficients

#1
I've run a logit regression to estimate the probability of attending high school. I've done this for many years. The thing is that some coefficients resulted to be not significant in some years and significant in some others.
So my questions are:

1- The fact that a coefficient is not significant means that the correspondent variable does not affect the probability I want to estimate?

2- How do I interpret the fact that one coefficient is significant at 1% level in one year and in the next year the same coefficient is not significant even at 10% level? Which should be my conclusion if in 5 of the 10 years I've considered the coefficients are not significant?

Thank you very much for your time!

PD: Sorry if my English is too bad, my language is Spanish.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
1- The fact that a coefficient is not significant means that the correspondent variable does not affect the probability I want to estimate? This means they are not statistically associated, which can get converted to a probability.


2- How do I interpret the fact that one coefficient is significant at 1% level in one year and in the next year the same coefficient is not significant even at 10% level? Which should be my conclusion if in 5 of the 10 years I've considered the coefficients are not significant? We cannot answer this for you, because we don't know these data or content. There could be a clear reason for this, but we have no idea without knowing the scenarios at play.
 
#3
Ok, thank you so much! I'll tell you more information of the data:

The sample includes all those people who have between 18 and 22 years old and have finished secondary school. The period under study are the 10 years between 2000 and 2010.
The dependent variable is a binary variable, which takes the value of 1 if the youth attend to high school and 0 otherwise.
The independents variables are:
Logarithm of family income
Parents education, which is represented by 3 binary variables: primary school, secondary school and high school; each of them tooks the value of 1 if at least one of the parents has finished the correspondent level of education, and 0 otherwise. The omitted variable is "no education".
Sex, which is a binary variable that takes the value of 1 if the youth is boy and 0 if it is girl.
Age, which is a control variable.

What I'm searching for is to determine if those factors that are beyond the control of the youth (sex, family income, etc.) are affecting his probability of attend high school.

But, for example, the coefficients corresponding to family income resulted to be significant one year, not significant at the next year, and so on. So, what can I say about the relationship between family income and the probability of attending high school?

Thank you very much again!
 

rogojel

TS Contributor
#4
hi,
could you check the SE of the coefficients for the different years? Do the confidence intervals for the coefficient values overlap? Maybe you have unequal sample sizes and in some years the sample sizes are not large enough?

regards
rogojel
 
#6
Hlsmith said...

"This means they are not statistically associated, which can get converted to a probability."

Are you implying that an insignificant coefficient does not affect the dependent variable?
 

hlsmith

Less is more. Stay pure. Stay poor.
#7
Without rereading the prior posts, I will hopefully answer this question. An insignificant coefficient can affect a dependent variable. However, it may not when controlling for other covariates and/or it may not meet the specified level of significance, but may explain part of the dependent variable. No idea if this answers your question or not.
 
#8
Without rereading the prior posts, I will hopefully answer this question. An insignificant coefficient can affect a dependent variable. However, it may not when controlling for other covariates and/or it may not meet the specified level of significance, but may explain part of the dependent variable. No idea if this answers your question or not.
OK. People often confuse insignificance with no effect and I wanted to see if that was how you felt about the matter.

I suppose the only thing I'd add is that most (if not all variables) will in some way correlate with the dependent variable, however small the effect size is.
 

noetsi

Fortran must die
#9
I interpret statistical signficance, as contasted with effect size, as how certain you are that the effect you found actually exists in the population. Something could have an effect in a sample, but not in the population due to random error (and the reverse is true, it might have an effect in the population, but not your sample due to low power for example).