Hi Svenja,

I'm attempting to do similar, although mine is linking pre/post-intervention responses rather than a diary study. 

I've used two questions: ‘what day of the month is your birthday?’ and ‘first three letters of your mother’s first name’.  So, if a respondent’s birthday was 1st January, 1970 and their mother’s first name was Catherine, their code would be 01cat.  

I've seen this type of approach in a few previous studies and included a few references below if you wanted to follow up.  But it seems to be reliable (Damrosch et al 1986) and it has been supported even when collecting potentially sensitive data (e.g. Garvey Wilson et al., 2010).  However, it's not perfect with losses of 20-30% due to missing or incomplete data not unusual (Schnell, Bachteler, and Reiher, 2010).

I have only collected the first phase of data, so can't tell you how effective it was for me yet - but from looking at that data I did find numerous duplicates in the sample (it looks like Mary or Margaret were VERY popular mothers names!).  I included a few additional demographic questions that enabled me to clearly differentiate them, so may be worth adding another question or including a demographic variable or two - depending on your sample.  Mine was 1400 so it was to be expected really.

Good luck,

John

Damrosch, S. P. (1986). Ensuring anonymity by use of subject-generated identification codes. Research in Nursing & Health, 9, 61-63.

Garvey Wilson, A. L., Hoge, C. W., McGurk, D., Thomas, J. L., Clark, J. C., & Castro, C. A. (2010). Application of a New Method for Linking Anonymous Survey Data in a Population of Soldiers Returning from Iraq. Annals of Epidemiology, 20(12), 931-938. doi: http://dx.doi.org/10.1016/j.annepidem.2010.08.008

Schnell, R., Bachteler, T., & Reiher, J. (2010). Improving the use of self-generated identification codes. Evaluation Review, 34(5), 391-418.

Yurek, L. A., Vasey, J., & Havens, D. S. (2008). The use of self-generated identification codes in longitudinal research. Evaluation Review, 32(5), 435-452. doi: 10.1177/0193841x08316676

Thanks a lot for this! Will have a look at the provided literature as well.

I have seen variations of the above method wherein Mother's Maiden names are used. This might help deal with overlap of more common names.

Here at UH we run quite a few diary studies and typically only use 3 digit, randomly-generated numbers. Based off of a dataset I cleaned last semester, I would wager that fewer than 200 out of 1200 responses are entered in error. Of those 200, many can be identified by IP addresses and dates in which the entries took place, which Qualtrics automatically records. 

We also tell participants that this code is how we know the correct amount of course credit to give them for their participation (a similar system should be manageable if you are providing payment), which incentivizes them to input it correctly.

If I am not mistaken, I think Qualtrics might be able to enter a code automatically for individual participants if you plan to email them links to the surveys.

Just as a follow-up to my earlier answer...I have now collected the 'time 2' data, so can tell you it worked ok, but I thought I would mention an issue that I hadn't anticipated.  Because anonymity was particularly important in my study I asked for first three letters of mother's first name as one of my matching code questions; I thought this would feel less identifiable to participants than asking about mother's maiden name, because the latter is often used in security questions for various things.  However, it was pretty clear that a lot of people had provided the first three letters of their mother's maiden name (e.g. McD, McC, Smi, etc...), for either one or both time points.  

On the one hand, perhaps it shows that people were willing to provide this info, but it looks like several people only half read the question.  I would use the same system again because it did pretty well, but I would definitely provide greater emphasis that I am asking about first three letters of mother's FIRST name.  Fortunately, I had a few other variables that I was able to use to match, but thought I would share my experience in case it helps anyone.

That is helpful indeed! John, thank you very much for sharing your experience!

This was very helpful! Thank you for the questions asked and the solutions given!

I have a sample of 800 participants who I'm tracking over time. From the lit it seems that accuracy reduces when the questions are about other people so I ended up using a code based on 5 questions:

1. First letter of your first name

2. Day of birth

3. Month of birth (which I then convert into a number)

4. First letter of your middle name (if none, use X)

5. First letter of city/town you were born in

So far, we've done 4 data collection waves from Sept to Jan and the accuracy rate is around 90% with no duplicate codes. Most people who make errors seem to do it on Q4 or 5. So overall, seems pretty successful!

Hi Svenja,

I am currently conducting my own longitudinal study. My first phase had 847 participants and now I am collecting second phase data. I can't tell you how well they match as yet but I tried to optimise this as much as I could. The papers mentioned below helped. I can keep you posted if you'd want that.

Three things are key as the papers below highlight:

a) Variability

b) Temporal stability

c) Proximal relevance

Another point is that you are asking for self-generated codes as you intend to keep information anonymous and confidential. Asking questions about the participant themselves (such as birth day and birth month) are usually not a great idea as it could appear are breaking anonymity.

The questions asked need to be clearly formulated and the person should know exactly what is being asked.. this is to ensure that there is not confusion when they create the code at a later phase of the study.

Example (statement – answer):

Month born: SEPTEMBER

Number of Older Brothers (half-brother, living, or deceased, if none write X): 1

Number of Older Sisters (half-sister, living or deceased, if none write X): X

First initial of Mother/Female Guardian’s First Name: S

First initial of Father/Male Guardian’s First Name: V

First initial of your middle name (if none, write X): X

Therefore, the unique code is (no spaces): SEPTEMBER1XSVX

References:

Cotter, R. B. et al. (2005) ‘Contacting participants for follow-up: How much effort is required to retain participants in longitudinal studies?’, Evaluation and Program Planning, 28(1), pp. 15–21. doi: 10.1016/j.evalprogplan.2004.10.002.

Diiorio, C. et al. (2000) ‘Focus on Research Methods An Evaluation of a Self- Generated Identification Code’, Research in Nursing & Health, 23, pp. 167–174.

Grube, J. W., Morgan, M. and Kearney, K. A. (1989) ‘Using self-generated identification codes to match questionnaires in panel studies of adolescent substance use’, Addictive Behaviors, 14(2), pp. 159–171. doi: 10.1016/0306-4603(89)90044-0.

Kearney, K. . et al. (1984) ‘Self-Generated Identification Codes for Anonymous Collection of Longitudinal Questionnaire Data’, The Public Opinion Quarterly, 48(1), pp. 370–378.

Kristjansson, A. L. et al. (2014) ‘Self-Generated Identification Codes in Longitudinal Prevention Research with Adolescents: A Pilot Study of Matched and Unmatched Subjects’, Prevention Science, 15(2), pp. 205–212. doi: 10.1007/s11121-013-0372-z.

Schnell, R., Bachteler, T. and Reiher, J. (2010) ‘Improving the use of self-generated identification codes.’, Evaluation review, 34(5), pp. 391–418. doi: 10.1177/0193841X10387576.

Yurek, L. A., Vasey, J. and Sullivan Havens, D. (2008) ‘The use of self-generated identification codes in longitudinal research’, Evaluation review, 32(5), pp. 435–52. doi: 10.1177/0193841X08316676.

These are great suggestions. I am also planning to link data between anonymous pre- and post-surveys that will be administered 3-4 months apart. Has anyone used the last 4 digits of the cell phone number, or does that number change too frequently?

I'm in agreement about most of this, however, in the US, birthday and birth month are considered protected health information and therefore probably SHOULDN'T be used.

I wonder about the sensitivity of using parent names (what if the participant has bad memories, or no memories of that person)?

I have used year of birth if the population is expected to be varied enough (this doesn't really work with school students)

I use two characters for the state of birth, but now run into people not born in the U.S. so use county or province as an alternative

I have used model year and two characters for make of first car, but now the younger adults in the U.S. may not be driving!

Not everyone has pets or siblings ...

I like the cell phone number - in the U.S., about 80% of people have cell phones now. Even if the number changes for some of them in that 3-4 months, if the other characters don't change, you're probably ok. But, with older populations, cell phone may be less widespread.

What about the first digits of one's original home address? Like if mine were 4567 Pine Street, use 45?

Can someone do the factorial and see how many digits we need for each population size?

I've thought about this a lot :)

I would avoid asking for mother's and father's initials because everybody does not have a mother and a father. In Norway, all teenagers have a cell phone number, so I will go for the first two letters of your birthday month + the last four digits of your cell phone number.

The use of names seems great, but when you are dealing with students from China or Korea, it gets a bit more complicated. The same goes for the use of 'first name' or 'last name' with extracted letters or initials as in some countries there are persons with a single name (not identified as 'first' or 'last'). The same goes for the name of the birth town, it may not be known in the latin script.... I have used the following 4 questions (for all students including internationals):

Please write down the first two letters of the name of your birth country (as it is called in English):

Please write down the day of the month on which you were born as a 2-digit number (e.g. 06, 12, 25):

Please write down your length in centimetres (as a 3-digit number) or inches (also as a 3-digit number) whichever system of measurement you are most comfortable with:

Please write down the day of the month on which your father was born as a 2-digit number (e.g. 01, 22, 30):

As this survey concerns adults their length is not going to change but they might not know this in either inches or centimetres so I let them answer this in either system, the effect for us is that this will increase the number of codes. The survey will collect in excess of 600 responses.

Any comments would be appreciated

Regards,

Robert

Hi,

Helpful answers everyone, thanks. We're trying to select the best code too but given we are looking at staff of an institution, personal information such as phone numbers and date of birth still felt too identifying espcially if matched with other answers ex: position, age. We thought of pincode but people might not want to do that - I think we'll use birthdays of parents, either one or a combination of both - though the point raised above that not everyone knows their parents is a valid one...

I‘m thinking the same thing, and it might be fine to replace unknowns with X. It is rare that people don’t know any parents at all, so there would probably not be that many full-X codes. Unless your sample is huge, but then you would need a longer code anyway. Plus, demographic data like sex or age (etc) can also help distinguish between double codes.

I am conducting a longitudinal study and I am confronted by the same challenge of identifying participants during follow-up. Thanks you for your answers.

With GDPR in the EU we have to be very careful about what information is being asked and why, so identifier that uses no personal information at all is very hard to design, but an example could be:

• Favourite historical person
• Favourite Disney character (just first name)
• The number of words in the title of your favourite book

So an example identifier will be Gandhi Mickey 2

This kind of questions will be generally temporally stable and will offer enough variability as a combination.

I have to say, these are only temporally stable if you can answer them at all. For me, I have either several or no answer to these particular questions, so the answers are bound to be subject to change. Or I will have to write down what I answered, which defeats the purpose of the SGID. I agree it’s not easy with data protection, but in my study, I avoided opinion or favourites questions for that reason.

I've also been thinking about this and I might ask participants for the first 2 letters of the place they were born (city, town, village) and the last 3 digits of their mobile phone number.

Good morning and the answers that everyone has provided are extremely helpful and exactly what I was looking for. However, I have a question about this process. My research with adolescent athletes and so we need to have a consent form which will contain the child's name and parents signatures. So I am assuming I can use any of this information when creating the code as this could be traced back to a individual? The consent forms will be kept separately from the questionnaires. But trying to get Ethical approval and need to get this 100% correct. Thanks for any advice. Jonathan

I have the same challenge. Originally used first two letters of mother's maiden name, age of mother, and last two digits of mother's birth year. Colleagues shared lack of a mother, a deceased mother, bad memories of mother, lack of international equivalencies to "maiden name." Use of any part of birthdates, street addresses, student ID numbers, phone numbers, may be traced back to the individual and considered Personal health info which is protected by law in the US. One suggestion was first two letters of favorite color. So many replies with suggestions from which to glean. Thank you, Everyone.

I have another issue. I am asking adolescents to fill in questionnaires and parents to fill in questionnaires and so I need the unique identifier to be the same (hopefully for both the parent and the child) so both need to be able to get it right and I am considering my questions for the unique identifier carefully. Any suggestions welcome.