Jaren Joe Valentino I think there's some confusion. A main effect simply refers to the association between 1 IV and the DV. If you have 4 independent variables and 1 dependent variable, there are 4 possible main effects.

From the sounds of it, your regression contains 2 significant main effects. To get a more detailed answer, please share the full output of the analysis.

Hello Jaren Joe Valentino

"1) The dependent variable (main effect) is insignificant"!!!

I think you are hiding something in your hand and you are not showing it.

Jaren Joe Valentino -

Don't worry about "significance." That changes with sample size. The relative significance between "independent" variables can change too, depending upon which ones you include, because they are not really "independent" in every way. (Consider collinearity, for example.)

What might really help is scatterplots. If you research the term "graphical residual analysis," that may help you a great deal. You can even compare the performances of different models. However, the results depend upon your sample, so if you can use more than one sample representing the population or subpopulation to which the model is to be applied, a separate scatterplot for each sample, you can compare model performances on each scatterplot, for each sample. You could also research the term "cross-validation."

You might want to consider "prediction" versus "explanation." The example in the appendix at the end of the following paper (open access version) may be helpful:

Galit Shmueli: 

 https://www.researchgate.net/publication/48178170_To_Explain_or_to_Predict.

Cheers - Jim

I agree with Blaine Tomkins that there appears some confusion, but I am assuming what you did is something different than what Blaine assumes. Though either of us could be write (or it might be something else. He says "A main effect simply refers to the [linear] association between 1 IV and the DV," which is correct (with the added word) if you conducted four separate regressions. I am assuming you ran a multiple regression using all four predictors (and since you talk about main effects, maybe you also had interactions) and in that can the main effect is conditional on the other effects. Cox and Donnelly (2011, p. 111) describe how the bad convention of saying the of effect of x1 on y (bad also for the word effect) when it should be x1|x2x3x4 in your case. That is if you ran a multiple regression. Anyway, you need to describe what you have done in more detail.

I read the question again, and as I originally read it, I still interpret it as simply confusing "main effect" with Y, and noting that his x-variables are predictors in a multiple regression. I think that the question is how can the dependent variable be "insignificant" if two of the predictors are "significant," and my reaction is to downplay "significance" and note that what you really care about is how well the prediction works with that combination of variables. For that purpose, visual aids (scatterplots here) can be very helpful, and I suggest graphical residual analyses, with cross-validation. (I also brought up the distinction between prediction and explanation and noted where an example can be found.)

Perhaps also of interest would be the variance of the prediction error, or other measures, if you want to know if a sample size is large enough. Don't forget heteroscedasticity.

[To more directly answer the question: Even if two predictors looked relatively good, that may not be enough. Anyway, I don't pay much attention to significance. Note that standard errors, like p-values, are sample size dependent, but standard errors are more readily interpretable.]

"1) The dependent variable (main effect) is insignificant " Are you referring to the intercept (not that you should be)?

What you have written is not at all clear.

Daniel Wright You raise an interesting point about specifying the distinction between a main effect of a predictor in a simple regression from a predictor included in a multiple regression since the two are not statistically equivalent. To which I have a follow-up question:

Suppose you conduct a multiple regression with predictors A and B, and you examine the estimate of predictor A. Although this estimate is influenced by B being included in the model, would one still not describe this as a main effect (i.e., simple effect)? If not, what (verbal) description is more appropriate?

Blaine Tomkins , this is the point Cox and Donnelly are talking about where it would be useful if packages output a notation like $\beta_{A|B}$ rather than just $\beta_A$. I think phrases like "simple effect" is so easily mis-interpreted and doesn't match onto its literal meaning that I avoid it.

Jaren Joe Valentino -

By "The dependent variable (main effect)..." and "main effect" later in your question, you really just meant "dependent variable," correct?

"Main effect" has a different meaning, and that confused people. I don't think that is what you meant. If you just say that you only meant "dependent variable," if that is what you meant, then that would clear up a lot of confusion.

So, if I am correct that you only meant "dependent variable," then please just say so. Otherwise, please explain.

Jim

Hi all,

Thank you for your input and for pushing me in the right direction - I appreciate this. Also, I apologize for the caused confusion, and I'll explain better.

I conducted an event study with over 500 events and calculated the cumulative abnormal return (CAR) for each event. The events are public announcements made by select firms. I then categorized each announcement into different technological capabilities, i.e., did the announcement contain technologies that mention “Entertainment”, “E-Commerce”, “Collaboration”, or “Education”.

For the regression, I used the CAR as the dependent variable, and the above-listed capabilities as the independent variables (coded as dummy variables). What I am trying to examine is if a firm makes such an announcement, will an abnormal return of their stock follow, and if yes, does the mentioned technological capability contained in the announcement contribute to this abnormal return?

I attached the regression output. If we consider model 14 containing all variables (dependent, independent, and control), would the following interpretation make sense: If a firm makes such an announcement, an abnormal return is likely to follow when the announcement contains technological capabilities in “E-Commerce”.

Thank you for your replies.

J.