As there is always some attrition in longitudinal research, then how can a researcher match the responses from same respondent at T1, T2 and T3...... in a longitudinal survey design? So that a researcher may carry out t-test later to check whether those who participated in all the surveys and those who left out earlier are significantly different.
In literature review, I found one of papers suggesting to ask for last 5 digits of cell phone number. But I have observed that different people do have matching last 5 digits in their cell phone numbers. So I am looking for other suggestions while keeping the anonymity of respondents.
If the data were designed with matching in mind, then you will have an identifier included. For example, I would in education and there will be a student identifier (and encrypted depending on who is using it and how it is being sent).
If you don't, then it becomes more difficult. If you have identifying information like names and addresses, then you can use approx. matching functions and try to find matches, but these values do change (names change, how people record ethnicity changes, phone numbers change, etc.), so you will get a number that describes how likely it is that two match. Depending on the use, this might be enough. We would often have to play detective with a small number of cases where the match was not clear.
You will need to have a data file that contains information how to contact people and you might have a separate data file that only holds identifiers and research data. You need to ensure the identifier are unique. I normally use a database to store the information on people and have other systems to collect the data. The database creates unique number to be used and these numbers can be used to connect to the data. SPSS (research data) and ACCESS (personal data) can be used to handle of this. It is very hard to get unique numbers. In the Netherlands I would like to use the BSN (civel status number), but I am no allowed to have it (I could get it, but am not allowed to). One unique identifier is the email address. At no point in time there will exist two identical email addresses. However I am not sure if this holds over time.
Reconstructing record or person identification is detective work and feasible for small sets of data, but very hard to do for large data sets. Tracking and tracing duplicates may take a lot of time and it will mess up you data.
Is the data already collected or are you still at the start of your project?
I was to face the same issue and going to consider following option (I think better put it with demographics at the end of questionnaire to allow respondents to consider your questions on variables first). Please note that I have not yet pilot tested it and will do it in Pakistan.
In Questionnaire:
Your 3 Alphabets: ________ Your 4 Digits__________
Instructions (very close to above information required)....
* Unique Code: You can make a unique code of your own to help us match your questionnaire at the second time as well. This code can be combination of any 3 alphabets and 4 digits in any sequence. You can write 3 alphabets and 4 digits and then combine in any way for example if I choose "Your 3 Alphabets" be A,M,N and "Your 4 Digits" be 1,3,6,9, I can combine them in any order to make a "Unique code" for example, A,6,9,M,3,N,1 or N,9,6,M,A,3,1 or any other way.
May be it can work. May be you can have more than 1 ways to match responses if the respondent even misses one.
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Applied Statistics
- Survey Design