Interpretation of the interaction term when the relevant dummy variable is insignificant

[Owen de Boer]

Dear all,

In the model that I have run to analyse the effect of currency swaps on the gross capital flows of the countries signing them I have included a dummy variable for the signing of such a currency swap agreement (signing=1) as well as a dummy for whether the country is a developed economy or a developing one (developing =1) To check whether the effect of a currency swap differs between a developing or developed country I have included an interaction term between these two dummy variables (signed and developing =1). However, the results that the model produced have rendered my interaction term significant at the 1% level with a positive coefficient, but my dummy variable for the signing of the currency swap is negative (as expected) but insignificant. How do I interpret these results? Due to the fact that the original currency swap dummy variable is insignificant it is impossible to conclude how large the positive effect of a currency swap is for a developing country right? However, does it still allow me to say that a positive relationship exist, but that the size of it is unclear due to the insignificance of the foregoing dummy variable? Many thanks in advance!

Kind regards,

Owen

[William Lisowski]

Welcome to Statalist.

Statalist members can better help you if we know what command you have run and what Stata gave you as output. Please review the Statalist FAQ linked to from the top of the page, as well as from the Advice on Posting link on the page you used to create your post. See especially sections 9-12 on how to best pose your question. It's particularly helpful to copy commands and output from your Stata Results window and paste them into your Statalist post using code delimiters [CODE] and [/CODE], as described in section 12 of the FAQ. Screenshots are often unhelpful and should be avoided.

The general advice is that considering the magnitude of the effect of an interacted variable is to use the margins command. In your case it might be something like

regress flow i.signiing##i.developing ...
margins swap

but others can explain it better than I.

For details, see the output of help margins and help factor variables (to understand the i.signiing##i.developing notation) and the full documentation for them linked to from the tops of the help output.

[Richard Williams]

The meaning of main effects changes once interaction terms are included. The fact that a main effect is insignificant may not mean much. See

https://www3.nd.edu/~rwilliam/stats2/l53.pdf

[julian mwanana]

Greetings all,
On my study i have generated 7 dummy variables representing seven 7 industry. On the table of correlation matrix majority of the dummies are strongly significant. So, i decided to include them into my regression model. Unfortunately, most of the dummies were dropped/omitted because of co-linearity problem. However i could wish to include them, how should i overcome this problem?

[Richard Williams]

Julian I am not sure you have a problem, or at least one that is solvable the way you want. Are the industry dummies mutually exclusive, i.e. you have to be coded 1 on one of them and zero on all the rest? Or can you have a score of one on more than one dummy?

If mutually exclusive you have to lose one dummy. But I don't know why you would lose most of them.

This handout may or may not help:

https://www3.nd.edu/~rwilliam/stats2/l11.pdf

If you tell us more about the coding of your variables we might be able to help more.

[julian mwanana]

Thanks prof. for a quick response. The dummy industries are mutually exclusive. I used this formula to generate them, so advice me whether its a correct way of doing it
first i have a column (sector) coding the industries as:
Financial = 1
consumer goods = 2
consumer services = 3
oil and gas = 4
industrial = 5
basic materials = 6
telecom = 7
utilities = 8
then i used this command to generate the dummies:
gen industrial =0
replace industrial = 1if sector = = 5

[Owen de Boer]

Julian I am not sure you have a problem, or at least one that is solvable the way you want. Are the industry dummies mutually exclusive, i.e. you have to be coded 1 on one of them and zero on all the rest? Or can you have a score of one on more than one dummy?

If mutually exclusive you have to lose one dummy. But I don't know why you would lose most of them.

This handout may or may not help:

https://www3.nd.edu/~rwilliam/stats2/l11.pdf

If you tell us more about the coding of your variables we might be able to help more.

Thank you for you response. The link clarified a lot!

[Richard Williams]

Thanks prof. for a quick response. The dummy industries are mutually exclusive. I used this formula to generate them, so advice me whether its a correct way of doing it
first i have a column (sector) coding the industries as:
Financial = 1
consumer goods = 2
consumer services = 3
oil and gas = 4
industrial = 5
basic materials = 6
telecom = 7
utilities = 8
then i used this command to generate the dummies:
gen industrial =0
replace industrial = 1if sector = = 5

Julian I think you've left out some details. As it stands you've only created one dummy, for industrial. You've said that most of your dummies got dropped, but without seeing their coding we can't say why. Maybe you accidentally created several dummies that had identical values instead of a separate dummy for each category.

In any event, you probably don't need to create dummies in the first place. Just do something like

reg y i.sector

See -help fvvarlist- for an explanation.