Skip to main content
Cross Validated

Return to Question

Added additional detail to help with providing help and answers
Source Link
K-J
K-J
  • 21
  • 3

I have similar recurring use case at work whereby we are wanting to understand attributes/features of a datasets such as 'how many customers talk about this', 'how many customers do that' and generally the only knowns that I think are relevant are the populatation sizes.

In this current example what we have is a dataset that containts 9636 survey responses and another dataset that has been previously analysed of customer feedback that has been categorised. The categorised dataset is different to the current survey dataset but we are running on the assumption that the categories will be similar. What we are seeking to understand is the proportions of the 9636 surveys that can be categorised into 1 of the 12 categories from the second dataset.

Is there a general sample size calculation method for this type of use case and if so can someone plese provide detail on what the method for calculating the sample size for this use case? I intend on calculating this in r so if anyone has any r code snippets that would help here I would also greatly appreciate it.

Thank you in advance! This is a recurring requirement at work (calculating sample sizes that will give our stakeholders confidence that the results are relevant to the population in question). As a side note, my stakeholders often ask for 'what sample size will be statistically significant' - could anyone additionally provide any feedback on technically whether this is the right question to be asking for the use case/s I have outlined in this post.

Thanks again. I appreciate any help or insight.

EDIT 1 - adding details to clarify Dataset 1 is a list of reasons why customers might need to speak to our contact centre and there is 1 category with 12 levels. Dataset 2 is the survey responses and we will classify the reason for low scores using the same 12 levels from dataset 1 (as there is an underlying assumption that the reasons for being unhappy about the product or service and thus result in a call to our contact centre, will be the same reasons why customers will state in a survey they are unhappy with the product or service.

EDIT 2 Dataset 2 is a set of survey responses that we need to categorise into (hopefully, if our assumptions hold up) 1 of 12 categories. Dataset 1 is where those 12 categories were initially defined but this was not a survey dataset - it was a dataset of transcripts from contact centre calls (and analysed them in order to categorise the 'root cause of their issue') so the customers in dataset 2 are unlikely for the most part to be the same customers in dataset 1.

EDIT 3 Essentially what we are trying to answer/test is 'are the reasons for customer calls the same reasons why customers are detractors in an NPS survey'. I am not sure what test I would run to test for this but my initial question is at the top of this process which is 'how do we calculate the size of the sample we need to analyse in dataset 2 in order to give confidence that whatever we find is likely to be representative of the whole population (of survey respondants).

I have similar recurring use case at work whereby we are wanting to understand attributes/features of a datasets such as 'how many customers talk about this', 'how many customers do that' and generally the only knowns that I think are relevant are the populatation sizes.

In this current example what we have is a dataset that containts 9636 survey responses and another dataset that has been previously analysed of customer feedback that has been categorised. The categorised dataset is different to the current survey dataset but we are running on the assumption that the categories will be similar. What we are seeking to understand is the proportions of the 9636 surveys that can be categorised into 1 of the 12 categories from the second dataset.

Is there a general sample size calculation method for this type of use case and if so can someone plese provide detail on what the method for calculating the sample size for this use case? I intend on calculating this in r so if anyone has any r code snippets that would help here I would also greatly appreciate it.

Thank you in advance! This is a recurring requirement at work (calculating sample sizes that will give our stakeholders confidence that the results are relevant to the population in question). As a side note, my stakeholders often ask for 'what sample size will be statistically significant' - could anyone additionally provide any feedback on technically whether this is the right question to be asking for the use case/s I have outlined in this post.

Thanks again. I appreciate any help or insight.

I have similar recurring use case at work whereby we are wanting to understand attributes/features of a datasets such as 'how many customers talk about this', 'how many customers do that' and generally the only knowns that I think are relevant are the populatation sizes.

In this current example what we have is a dataset that containts 9636 survey responses and another dataset that has been previously analysed of customer feedback that has been categorised. The categorised dataset is different to the current survey dataset but we are running on the assumption that the categories will be similar. What we are seeking to understand is the proportions of the 9636 surveys that can be categorised into 1 of the 12 categories from the second dataset.

Is there a general sample size calculation method for this type of use case and if so can someone plese provide detail on what the method for calculating the sample size for this use case? I intend on calculating this in r so if anyone has any r code snippets that would help here I would also greatly appreciate it.

Thank you in advance! This is a recurring requirement at work (calculating sample sizes that will give our stakeholders confidence that the results are relevant to the population in question). As a side note, my stakeholders often ask for 'what sample size will be statistically significant' - could anyone additionally provide any feedback on technically whether this is the right question to be asking for the use case/s I have outlined in this post.

Thanks again. I appreciate any help or insight.

EDIT 1 - adding details to clarify Dataset 1 is a list of reasons why customers might need to speak to our contact centre and there is 1 category with 12 levels. Dataset 2 is the survey responses and we will classify the reason for low scores using the same 12 levels from dataset 1 (as there is an underlying assumption that the reasons for being unhappy about the product or service and thus result in a call to our contact centre, will be the same reasons why customers will state in a survey they are unhappy with the product or service.

EDIT 2 Dataset 2 is a set of survey responses that we need to categorise into (hopefully, if our assumptions hold up) 1 of 12 categories. Dataset 1 is where those 12 categories were initially defined but this was not a survey dataset - it was a dataset of transcripts from contact centre calls (and analysed them in order to categorise the 'root cause of their issue') so the customers in dataset 2 are unlikely for the most part to be the same customers in dataset 1.

EDIT 3 Essentially what we are trying to answer/test is 'are the reasons for customer calls the same reasons why customers are detractors in an NPS survey'. I am not sure what test I would run to test for this but my initial question is at the top of this process which is 'how do we calculate the size of the sample we need to analyse in dataset 2 in order to give confidence that whatever we find is likely to be representative of the whole population (of survey respondants).

Source Link
K-J
K-J
  • 21
  • 3

Calculating Sample Size When Wanting To Analyse Proportions & Proportions Sizes are Unknown

I have similar recurring use case at work whereby we are wanting to understand attributes/features of a datasets such as 'how many customers talk about this', 'how many customers do that' and generally the only knowns that I think are relevant are the populatation sizes.

In this current example what we have is a dataset that containts 9636 survey responses and another dataset that has been previously analysed of customer feedback that has been categorised. The categorised dataset is different to the current survey dataset but we are running on the assumption that the categories will be similar. What we are seeking to understand is the proportions of the 9636 surveys that can be categorised into 1 of the 12 categories from the second dataset.

Is there a general sample size calculation method for this type of use case and if so can someone plese provide detail on what the method for calculating the sample size for this use case? I intend on calculating this in r so if anyone has any r code snippets that would help here I would also greatly appreciate it.

Thank you in advance! This is a recurring requirement at work (calculating sample sizes that will give our stakeholders confidence that the results are relevant to the population in question). As a side note, my stakeholders often ask for 'what sample size will be statistically significant' - could anyone additionally provide any feedback on technically whether this is the right question to be asking for the use case/s I have outlined in this post.

Thanks again. I appreciate any help or insight.