Go through this link

https://www.bing.com/ck/a?!&&p=dd38d0fac7ad5e79JmltdHM9MTcwNzYwOTYwMCZpZ3VpZD0xMzJmMWQ5ZS05NDA4LTY1ZjgtMTIyYS0wZWI1OTUxNTY0YTEmaW5zaWQ9NTE5NA&ptn=3&ver=2&hsh=3&fclid=132f1d9e-9408-65f8-122a-0eb5951564a1&psq=Does+adding+an+time+trend+variable+in+a+fixed+effects+model+make+sense%3f+&u=a1aHR0cHM6Ly9zdGF0cy5zdGFja2V4Y2hhbmdlLmNvbS9xdWVzdGlvbnMvMzMwNDkyL2Nhbi10aW1lLXRyZW5kLXZhcmlhYmxlcy1iZS1maXhlZC1lZmZlY3Rz&ntb=1

hey Chuck A Arize , thanks but i don't think that thread answers my question

Either include the event-variable (0=before, 1=after) or the time variable (continuous,

Hi Dr Jochen Wilhelm ,

thanks for your input. But i have seen several papers that have included both event and time variable. For example look at this paperArticle Rallying around the EU flag: Russia's invasion of Ukraine an...

In addition to what Jochen said, it is possible for autocorrelation to appear.

You can consider time in another way, such as the time elapsed between events - this is because the perception changes as a function of the number of events in a row,

Therefore, the more events in a short period, the more the perception should change in the face of a low frequency.

Gjergji Josifi , the authors use normal regression on survey response, t-tests to compare frequencies between groups, t-tests in descriptive statistics, and they claim that the regression coefficients indicate causal effects in models based on survey data. That makes me skeptic about the quality of the analysis.

It's hard to judge from what is provided (I checked the employments of the authors via ORCHiD), but as far as I can see there is no statistician involved in the whole study. I also don't think that the manuscript has been peer reviewed by a statistician.

The regression model included a linear trend. No investigation or plausibility check was done that this would be a rational choice. Given the "hit & impact" character of the intervention, a linear relationship is highly implausible to my (very limited) understanding. A broken-line or segmented regression model might have been more useful (still not assuming any linear trend; I'd keep that more descriptive and use a GAM).

And the most important thing: as you can see, the CIs in "full model" are systematically wider than those of the "intermediate model", what one would expect to see when I was correct in my previous post.

Jochen Wilhelm I see your point, but i have seen numerous other papers that do include both event and time variable. Here is another one: Article Will you still trust me tomorrow? The causal effect of terro...

My only answer is: it does partially ruin this time dummy variable.

the answer to your question is Yes,

Consider the following when reading the papers,

will you believe a model that tells you what you want to read, even though you don't know its validity?