Hello,
I have posted a related question that refers to obtaining unexpected results and might have overlooked an important fact. I am conducting an analysis on how certain psychological characteristics of CEOs influence their speech during quarterly conference calls.
My unit of analysis is firm-quarter. For a sample of S&P 500 companies, I have gathered transcripts of CEO speeches from quarterly conference calls spanning 2010 to 2018. The dependent variables in my study include textual attributes such as the positivity of speech. That is, each observation relates to a conference call of a firm - or the linguistic attributes of a CEO's speech during the call.
I have several control variables, most of which are also measured quarterly. These controls include factors like net income reported by the firm in a given quarter (as one would expect that CEOs use more positive speech when the quarter went well). Other control variables are time-invariant, such as the firm's industry classification or a CEO's gender.
The primary variable of interest is the CEO's psychological trait of narcissism, which is proxied using multiple variables aggregated into a single proxy - using either factor analysis or combining the averages of the standardized items.
Here's the challenge: the variables used to construct the overall narcissism proxy are measured at different frequencies. For instance, one variable reflects the number of lines in a CEO's biography, which remains constant throughout their tenure. Conversely, other variables which are part of the narcissism proxy, like CEO compensation, are measured annually.
In a nutshell, I have a panel dataset in which the DV is measured quarterly while my IVs are collected on either an annual level, quarterly level or remain constant across time.
I have reviewed various articles on this topic but haven't found a suitable solution for my specific scenario. Am I correct in assuming that fixed effect regression would be influenced by these variations in periodicity? Any ideas on how to handle such data structures would be greatly appreciated.
Thank you
I have posted a related question that refers to obtaining unexpected results and might have overlooked an important fact. I am conducting an analysis on how certain psychological characteristics of CEOs influence their speech during quarterly conference calls.
My unit of analysis is firm-quarter. For a sample of S&P 500 companies, I have gathered transcripts of CEO speeches from quarterly conference calls spanning 2010 to 2018. The dependent variables in my study include textual attributes such as the positivity of speech. That is, each observation relates to a conference call of a firm - or the linguistic attributes of a CEO's speech during the call.
I have several control variables, most of which are also measured quarterly. These controls include factors like net income reported by the firm in a given quarter (as one would expect that CEOs use more positive speech when the quarter went well). Other control variables are time-invariant, such as the firm's industry classification or a CEO's gender.
The primary variable of interest is the CEO's psychological trait of narcissism, which is proxied using multiple variables aggregated into a single proxy - using either factor analysis or combining the averages of the standardized items.
Here's the challenge: the variables used to construct the overall narcissism proxy are measured at different frequencies. For instance, one variable reflects the number of lines in a CEO's biography, which remains constant throughout their tenure. Conversely, other variables which are part of the narcissism proxy, like CEO compensation, are measured annually.
In a nutshell, I have a panel dataset in which the DV is measured quarterly while my IVs are collected on either an annual level, quarterly level or remain constant across time.
I have reviewed various articles on this topic but haven't found a suitable solution for my specific scenario. Am I correct in assuming that fixed effect regression would be influenced by these variations in periodicity? Any ideas on how to handle such data structures would be greatly appreciated.
Thank you
Comment