Can you please explain me what is SN? I think dropping one may cause measurement error in the variable which brings biased results.

Mohsin

Hello Hasan Mohsin,

SN = subjective norms 

I wish I could be of more use but, I'm interested in the answer to this question as well.

Daryn

Yes, Thanks Dayran Dyer. SN is subjective norm. We have the two questions for direct SN:

Q21: On a scale from 1-7 where 1 = Not Should, and 7 Should, please rate the following statement. "People I care about think that i should not/should buy organic baby food. "

and

Q27: On a scale from 1-7, where 1 = Very Unlikely and 7 = Very Likely, please answer the following question. "How likely is it that you will listen to advice from people important to you when you buy organic baby food?"

The two questions has a low degree of internal consitency, but which question do we keep for our multiple regression analysis?

So both the SN measures, 7 means very likely to buy organic food and 1 means less likely. Now tow terms are used here which are interesting,. 1. People i care about  2. people important to you. 

I will say use both the measures one by one and check results. If results are same with both measures one can argue that i have used as robustness measure. If the results are opposite then i will raise a question regarding how question is framed.

People i care about seems emotional attachment. I care about my mother and wife but their advice may not be always true about food (organic or inorganic). But the advice of doctor, nutritionist or scientist is more accurate but do i care about them or are those important to me. Please share your thoughts, if it is interesting point

go for pilot survey .. you will get the ans

The only way to have a low degree of internal inconstancy with two items is to have a low correlation between them. The first thing you should do in that case is to check for problems due to missing data etc. by examining a scatter plot.

If the two questions are indeed uncorrelated, then they have tapped into completely different things. I would then recommend treating they as two different aspects of SN and retaining both of them in your analyses. In particular, if you are running regressions, they each have the potential to make a separate contribution precisely because they are uncorrelated.

Q27 seems confusing to me, and I can see how it might be problematic. The end of the question, "when you buy organic baby food?," suggests that the decision to buy organic baby food has already been made, thus why listen to someone's advice?  It seems like perhaps a word is incorrect or missing (?) If your questions were translated from one language to another, double check to make sure the translations are accurate - a missing or inaccurate word can change the intended meaning, and thus contribute to low internal consistency. Also, see this post regarding internal consistency reliability for two-item scales:  https://www.researchgate.net/publication/232610246_The_reliability_of_a_two-item_scale_Pearson_Cronbach_or_Spearman-Brown

Article The reliability of a two-item scale: Pearson, Cronbach, or S...

Thanks for your inputs. Alison West, i can see that i have a mistake in Q27, it should state: (Minus the organic in front of baby food).

Q27: On a scale from 1-7, where 1 = Very Unlikely and 7 = Very Likely, please answer the following question. "How likely is it that you will listen to advice from people important to you when you buy baby food?"

However, we followed the article you posed and did a cronbach alfa and a pearson test and found low cronbachs (0,2) and no significance for the Pearson test. So, probably the questions should have been stated in a different manner when we constructed the questionnaire.

I think we will include both the measures as David Morgan suggested. Thanks.

However, the same thing happened for PBC. Do we keep both in this direct measure as well?