How can I treat my pilot data that has 1 repeated indicator/survey item for all 3 companies in R Studio?

Question

I'm conducting a study on corporate social responsibility (CSR) and am encountering a challenge with my data cleaning. I've instructed respondents to answer the Likert scale questions only for companies they are familiar with. All 29 indicators/survey items are repeated per company (There are 3). However, I've noticed that three respondents have marked "N/A" for all 29 indicators of two companies.

My concern is that this could lead to a significant number of "N/A" responses, which might be interpreted as zeros in R Studio, potentially affecting my statistical analysis in the actual data collection of atleast 400 respondents.

Given the repetition of the same indicator for each company, I'm wondering if there are effective strategies to handle data in this context.

1. I've received a recommendation to count each company response as count 1 irregardless if it is under 1 respondent. (1st Respondent Company 1 = 1st Response, Company 2 = 2nd Response). However, the demographic profile as the Moderating Variable will be skewed, right?

2. I've also received a recommendation to present 3 results per company. Won't this become a case study instead of an academic paper?

Answer 1

First off, R will not handle the missing values as zeroes. It will however potentially bias your results for different reasons. As a basic example, R will by default simply kick rows out of your data which include missing values, but this will only happen if they are included as-such in your data. So as your post suggests, if these are aggregated summaries, then this would only affect the aggregated scores if they are completely missing. Though if people consistently do not answer on those questions for other companies, then this is still something left unaccounted for in your model.

Something you could do here is treat this as an imputation problem. If we know the responses of others on the same items, we can essentially use that information to predict responses of others. This could be done using something like multiple imputation. Since you are using R, the mice package is good for this (see vignettes for the package here). This would ensure that there are no missing values in your analysis, you get to retain your data, and the imputed data provides the most plausible "guesses" of what those values should be. Much research has shown that principled versions of imputation like this are far superior to simply omitting the missing values together or using other half-baked solutions.