I have a very basic question about interpreting confidence intervals. Say I use R to fit a linear model to the Nile water flow time series (from the datasets package), like this:

# Flow of the Nile river (10^8 cubic metres per year) between 1871-1971
nile.ts <- data.frame(year = do.call(seq, as.list(attributes(Nile)$tsp)),
           flow = as.numeric(Nile))

# Fit linear function 
nile.fit <- gls(flow ~ year, data = nile.ts)

Next, I look at the confidence intervals on the model parameters:

# Look at 95% confidence intervals
Approximate 95% confidence intervals

                  lower        est.       upper
(Intercept) 4144.217893 6132.173579 8120.129266
year          -3.749313   -2.714305   -1.679298
[1] "Coefficients:"

 Residual standard error:
   lower     est.    upper 
132.1041 150.5522 175.0363

Here, the coefficient of year in the model is estimated to be $-2.7$ and has a 95% confidence interval of $[-3.7, -1.7]$. I'd read this as meaning I can be pretty confident that there has been a decline in the flow over time. But what if the confidence interval had been $[-3.7, +1.7]$? Since it straddles zero, does this mean that I can't say anything about the trend? Or, can I state that there is no statistically significant trend?


1 Answer 1


Or, can I state that there is no statistically significant trend?

Yes. The coefficient for year would be then not different from 0.

  • $\begingroup$ @ Lyngbakr, you can use a more formal wording that is based on the p-value in the model output, which is that: the probability that a coefficient estimate will be at least as large as the observed value when the Null hypothesis is true. $\endgroup$ Oct 26, 2017 at 13:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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