You actually have the numbers of births (per1000). That's a count variable and should thus be analyzed with a count model (Poisson, negative-binomial, or a quasi-model),possibly a zero-inflated count model.

You actually have the numbers of births (per1000). That's a count variable and should thus be analyzed with a count model (Poisson, negative-binomial, or a quasi-model),possibly a zero-inflated count model.

You can do beta regression. The trick is to use the mean-variance parameterization of the beta distribution. If you don't want to implement yourself, there is some existing R package that can do it. I am pretty sure it's also available in other software.

Here is a link to some advice. I like the limited dependent variable approach myself. I t is often used in econometrics.. Best wishes, David Booth

https://www.theanalysisfactor.com/proportions-as-dependent-variable-in-regression-which-type-of-model/

If your dependent variable goes from 0 to 1 (real values), then the Beta regression is best suited for your work because.

If it takes integer values (count data), then the count models are best suited.

Is this a life table question, or are you only using the birth rate? The birth rate is only one of several demographic parameters, but you would also need age specific mortality rates to go further.

An alternative might be to fit a logit regression. If your dependent variable is p which varies between 0 and 1 (or P which varies between 0 and 100) and x is your independent variable, calculate the logit transform of the dependent variable as y=log{p/(1-p)} or y=log{P/(100-P)} using natural logs. Plot a scatter plot of y vs x. If it looks like the scatter plot of a linear regression, then go ahead and fit a straight line of y on x, say y=a+bx. To transform the model from y to p, calculate p={exp(a+bx)}/{1+exp(a+bx)} or replace p by P and the 1 by 100 if you have percentages. The resulting curve of p vs x has an S shape. A similar curve to try could be the probit curve. SPSS has a routine Analyse-Regression-Probit and then you have a choice between logit and probit.

.

Jochen Wilhelm is right : if the DV is indeed the number of birth per 1000 women, it is a count (could be anything from 0 to 20000 !) and not a proportion !

(and the question is poorly phrased as it states "proportion" in the title)

.

Now, if the DV is something like the number of women having given birth (at least once ...) in 1000 women, then this would be a proportion (could only take values in from 0 to 1000) and not a count but this seems unlikely from the question

(and the question should be clarified)

.

maybe the question could be rephrased to be less ambiguous

.

try ADL or ECM model.

A good option will be to use logistic regression. Since the logit transformation will

ensure that the predicted value will lie in the range of 0 to 1.

For dependent variable from the [0, 1] interval you can apply the fractional response model developed by Papke and Wooldridge (1996), see: https://onlinelibrary.wiley.com/doi/pdf/10.1002/%28SICI%291099-1255%28199611%2911%3A6%3C619%3A%3AAID-JAE418%3E3.0.CO%3B2-1

For dependent variable from the (0, 1) interval you can use the beta regression approach proposed by Ferrari and Cribari-Neto ( 2004 ), see:

Article Beta Regression or Modeling Rates and Proportions

Regards,

Hanna

Thank you for your valuable suggestion for know I used the Poisson regression taking number of birth as the dependent variable to see the indicator of birth in national and sub-national level using PDHS data sets.

I think logistic regression will be more applicable in your case since your dependent variable isn't continous