x
Facilitating low literacy among debtors from the point
of view of a debt collection agency
Master thesis
Emre Kumru
Groningen, June 17
th2019
Abstract
About 18% of the population of the Netherlands is low literate, and this costs the Netherlands around €1.13 billion on a yearly basis (Stichting Lezen & Schrijven and PwC, 2018). This research project looks into ways to facilitate low literacy from the view of debt collection agencies.
The main purpose of this research is to analyze the effects of simplified reminders for debtors. The purpose of this research led to the research question: ‘What type of debt reminder is the
most effective for debtors?’
More than ten thousand various reminders have been sent to debtors during the sampling period. The chosen debtors in this research had an outstanding debt at a Dutch public transport company. Afterwards, effectiveness of simplified reminders has been assessed in this research. The effectiveness of the different treatments is assessed by making use of binomial logit models. All binomial logit models had a hit-rate of 80.5% or higher which indicate reliable results.
The main findings of this research are that the usage of simplified text and the usage of pictograms in reminders seem to be ineffective compared to the original reminder which makes use of long sentences, uses jargon and does not make use of pictograms. This research recommends debt collection agencies to classify each debtor into meaningful groups based on relevant variables. Targeting each debtor appropriately should increase compliance among debtors.
Colophon
This document is a Master thesis for the completion of the Master Marketing
Intelligence at the University of Groningen, The Netherlands.
Title:
Facilitating low literacy among debtors from the
point of view of a debt collection agency
Version:
Final version
Date:
03-06-2019
Author:
Emre Kumru
Student number:
S3800024
Address:
Het Hout 125, 9723 LB Groningen (NL)
Phone number:
+31651986276
E-mail address:
E.Kumru@student.rug.nl
University:
University of Groningen
Faculty of Economics & Business
Department of Marketing
PO Box 800, 9700 AV Groningen (NL)
Master Marketing (Intelligence track)
First supervisor:
dr. Martijn Keizer
Second supervisor:
dr. Jelle Bouma
Data received from:
Syncasso Nederland B.V.
Preface
Dear reader,
This thesis is made by Emre Kumru. This research project finalizes my
academic career at Rijksuniversiteit Groningen. In this preface, I want to thank
some people for their help and support.
First, I want to thank my supervisor Martijn Keizer for his critical academic
view on my thesis and his relevant feedback.
Secondly, I want to thank my second supervisor Jelle Bouma who connected me
with Martijn Keizer and gave me helpful tips at the start of my thesis process.
Finally, I want to thank my friends, parents and siblings who always motivated
and supported me during my study.
Hopefully you will take pleasure in reading this thesis.
Kind regards,
Table of contents
ABSTRACT ... 2 COLOPHON ... 3 PREFACE ... 4 1. INTRODUCTION ... 6 1.1PREVIOUS RESEARCH ... 81.2RESEARCH QUESTION AND SUB-QUESTIONS ... 9
1.3OUTLINE ... 10
2. THEORETICAL FRAMEWORK ... 12
2.1LITERACY ... 12
2.2READABILITY AND LEGIBILITY ... 13
2.3DEBT LITERACY ... 15 2.4CONCEPTUAL MODEL ... 17 3. METHODS ... 18 3.1RESEARCH DESIGN ... 18 3.2MANIPULATIONS ... 18 3.3DATA ... 19 3.3.1 Sample ... 20 3.4LOGISTIC REGRESSION ... 21
3.5DEPENDENT VARIABLE & INDEPENDENT VARIABLES ... 21
4. RESULTS ... 24
4.1DATA CLEANING ... 24
4.2DESCRIPTIVE STATISTICS ... 25
4.3BINARY LOGISTIC REGRESSION ... 26
4.3.1 Binary logistic regression assumptions ... 28
4.3.2 Model fit ... 29
4.4RESULTS HYPOTHESES ... 30
4.4.1 The effect of simplified text and usage of pictograms in the first debt reminder ... 30
4.4.2 The effect of simplified text and usage of pictograms in the second debt reminder ... 31
4.4.3 The effect of amount of a debt on the effectiveness simplified reminders ... 31
4.4.4 The effect of a debtor living on a risky address on paying a debt. ... 32
4.4.5 The effect of age on the effectiveness of simplified reminders ... 33
4.4.6 The effectiveness of the type of reminder per sub-group for reminder 1 and reminder 2 ... 33
4.5SUMMARY HYPOTHESES AND RESULTS ... 35
5. CONCLUSIONS ... 37
5.1CONCEPTUAL MODEL RESULTS ... 37
5.2MAIN FINDINGS ... 37
5.2.1 Answering sub-questions ... 38
5.3DISCUSSION FINDINGS ... 40
6. RECOMMENDATIONS ... 42
6.1RECOMMENDATIONS FOR DEBT COLLECTION AGENCIES ... 42
6.2RECOMMENDATIONS FOR POLICY MAKERS ... 42
6.3LIMITATIONS ... 43
7. REFERENCES ... 45
8. APPENDICES ... 47
1. Introduction
Even in these digital and modern times, a western country based on a knowledge economy like The Netherlands is still populated with a fairly high amount of low literate people. Out of the population with an age of 16 years or older in the Netherlands, 2.5 million of them are low literate (Stichting Lezen & Schrijven, 2012); this is about 18% of the population of the Netherlands. Furthermore, 1.9 million people older than the age of 16 have difficulties with reading and do not possess of basic mathematical skills. A low literate person in the Netherlands is defined as a person who is struggling with reading, writing or calculating and when these possessed skills are below the level of a VMBO (preparatory secondary vocational education) graduate.
Approximately, low literacy costs the Netherlands around €1.13 billion on a yearly basis (Stichting Lezen & Schrijven & PwC, 2018). The major part of the calculated costs (51%) are allocated to the low literate persons themselves, but the rest of the costs are allocated to tax payers which makes it a problem for the whole society (PwC, 2018). For example, because of a part of the Dutch population does not pay their health insurance bills in time (or not at all), health insurance companies have to raise the premiums to make up for the incurred costs of non-paid health insurance bills. These raises in health insurance premiums are at the expense of paying Dutch residents. Additionally, low literate persons often make use of debt assistance services offered by municipalities in the Netherlands (Rijksoverheid, 2017). Costs incurred due to low literacy arisen from debt assistance offered by municipalities in the Netherlands costs the tax payer an estimated €238 million (Rijksoverheid, 2017).
literate persons in order to reduce poverty. Education programs can help low literate people to be more educated and thereby spend their money in smarter ways and also try to save money.
Furthermore, in the field of health literacy, it has been proven that communicating in a simpler language is enhancing patient compliance (Mayeaux et al., 1996). Low literate persons often have a hard time understanding health instructions (Mayeaux et al., 1996). Failing to follow the debt collectors’ instructions as a low literate person and eventually being worse off is similar to research about compliance in the health sector where low literate persons fail to understand the health enhancing instructions and do not comply with the instructions and are eventually worse off. Academic papers also study how to facilitate low literate persons, therefore, research about health compliance can be linked to compliance targeted at debtors.
Lastly, Lusardi & Mitchelli (2007) show that education programs which are aimed to reduce financial illiteracy and have a ‘one-size-fits-all’ approach fail to be effective. They advise policy makers to organize different educational programs aimed at different sub-groups. This research project will implement the main thoughts and results of the previous mentioned papers, this research project will address the topic of enhancing debtors’ compliance with help of making instructions clearer on how to pay off debts, while trying to target various possible sub-groups in the society effectively.
This research project will try to analyze the effects of facilitating low literacy by communicating with simplified texts and/or using pictograms to low literate people from debt collection agencies. More specific, this research project will try to aim at different sub-groups in different ways with help of a binomial logistic regression model in order to effectively reduce debts for low literate persons and as a result of this reduce costs for the whole society.
example, when the debt collection agency would sue a debtor without chance of success (Keizer, 2019).
1.1 Previous research
Illiteracy can be approached in different ways; illiteracy can be reduced with use of educational programs as an intervention. This way, low literate persons can be taught how to read, write or calculate in a person’s important personal financial environment. Another way to approach illiteracy is to make use of interventions to facilitate illiteracy. For example, communicating in a way that a low literate person understands instructions in order to enhance compliance.
Previous research on literacy is conducted on both reducing literacy and facilitating illiteracy in the field of health literacy and financial literacy. Previous research shows that reducing illiteracy is mainly done through educational programs (Kirsch, 1992; Lusardi & Mitchelli, 2007; Huston, 2010). Facilitating illiteracy is mainly done by making instructions clearer for the low literate person. Both ways of approaching illiteracy can be researched, but this research project will focus on interventions on how to facilitate illiteracy, in particular, for collection agencies. Mainly, how to communicate toward low literate persons and increase compliance levels through simplified forms of communication will be researched. For example, Mayeaux et al. (1996) show that simplified forms of communication are helpful in ways to enhance patient compliance in healthcare. This approach is chosen, because from the perspective of a debt collection agency it is appreciative knowing how to handle communication toward low literate persons. Another advantage from facilitating low literate debtors is that it is cheaper to facilitate low literate debtors than to reduce low literacy among debtors due to the high cost and effort necessary when training low literate persons and thereby reducing low literacy among these people.
pictograms can differ between reminders. Concerning the usage of pictograms and simplified text for better communication toward low literate people, Plimpton & Root (1994) find that the core message must be clear, appropriate language (no technical language or jargon) must be present next to illustrations fitting the message. The effects on compliance for usage of pictograms and simplified text will also be analyzed in this research project.
The reminders in this research project will consist of four different reminders each differing in the level of text difficulty and the usage of pictograms. The control reminder is the version of the reminder a Dutch debt collection agency was sending out previously before and during this research project. This reminder is a version with relatively long sentences, using jargon and without using pictograms supporting the message. This type of reminder is being used, because the debt collection agency wants to include all the relevant information and needs to conform to the rules of juridical way of communicating. But, because a large part of debtors at this Dutch debt collection agency (59.6%) is considered low literate (Keizer, 2019), there needs to be researched whether higher compliance can be achieved by communicating differently.
Moreover, this research project will shine light on which sub-groups to target in what manner. Every debtor is classified in a sub-group on the basis of relevant variables. The sub-groups will be used to assess effects of each reminder for every sub-group. Thereafter, each sub-group out of debtors can be targeted effectively. The process of classifying each debtor into a group is further explained in chapter 3. Furthermore, previous research in this field of area is mostly done in the U.S.A., this research project will solely include data from the Netherlands, and will try to add scientific knowledge in this field of area for the Netherlands.
1.2 Research question and sub-questions
The introduction and previous research bring us to the research question of this thesis.
This research project will focus on making tweaks in the communication toward debtors living in the Netherlands and assessing the effects hereof. The main communication toward debtors will be done using one page long letters reminding debtors to pay their debt off.
This research project will assess the effects of different tweaks in reminders toward debtors and will afterwards classify each debtor into a group where he or she will get the most effective type of reminder in their mailbox. An effective reminder causes a debtor to take action, for example by paying their debt, or by contacting the debt collector when a debtor is unable to pay their debts. However, the contents of mails and telephone calls between a debtor and debt collection agency are not registered and can, therefore, not be used as a variable to assess the effectiveness of a debt reminder. This research project will only look at the effectiveness in terms of payments.
In this research project, every debtor receives one or two reminders wherefrom can be assessed which effects play a role for an effective reminder.
Several sub-questions are necessary in order to answer the main research question.
1. Is the usage of fitting pictograms and a simplified text an effective way in order to make the debtors pay off their debt?
2. For the group of debtors who received two reminders, which second reminder was the most effective?
3. Does the effectiveness of a reminder change between different amounts of debts? 4. How does the living situation affect the effectiveness of each reminder?
1.3 Outline
2. Theoretical framework
2.1 Literacy
Literacy is one of the main topics of this research project. In the next three paragraphs, various points of view toward literacy will be discussed. Afterwards, there will be explained how this term will be used throughout this research project.
The term ‘Literacy’ is defined - in the Netherlands – as following: “Low literacy is a term for people who struggle with reading, writing and/or calculating. People who are low literate are not illiterate; low literate people can read and write, but do not possess of the literacy level of a VMBO (Dutch preparatory secondary vocational education) graduate.” (Stichting Lezen & Schrijven, 2019). The Cambridge dictionary (2019) defines ‘Illiteracy’ on the other hand as: “The fact of being unable to read and write”. This research project will focus on low literate people, because effects of facilitating low literacy by using simplified reminders can only be tested on low literate persons out of illiterate and low literate persons. Illiterate persons are not able to read and therefore the effects of the different reminders cannot be tested.
Literacy in health related research is defined as: "The degree to which individuals have the capacity to obtain, process, and understand basic health information and services needed to make appropriate health decisions." (Ratzan & Parker, 2000).
Financial literacy is defined as: “The process by which financial consumers/investors improve their understanding of financial products and concepts and, through information, instruction, and/or objective advice, develop the skills and confidence to become more aware of financial risks and opportunities to make informed choices, to know where to go for help, and to take other effective actions to improve their financial well-being.” (OECD, 2005)
seen as somebody who is unable to fully understand and follow the instructions given by the debt collection agency.
As earlier stated, low literacy can be facilitated by simplified instructions (Mayeaux et al., 1996). Consequently, the process of constructing effective instructions call for more explanation. The relevant aspects of creating instructions such as the difficulty and layout of a text are discussed in the next section.
2.2 Readability and legibility
Readability is the level of difficulty of a text (DuBay, 2004). The main guidelines for readability are compiled from golden rules of documentation writing and are as follows (DuBay, 2004):
• Texts should consist out of short, simple and familiar words.
• Texts should avoid jargon, and should consist out of culture-and-gender-neutral language.
• Texts should use correct grammar, punctuation and spelling.
• Texts should consist of simple sentences, active voice, present tense.
• Begin instructions in the imperative mode by starting each sentence with an action verb. • Use simple graphic elements to support your text
Mcinnes & Haglund (2011) show in their scientific paper that readability is an integral part in facilitating low literacy in health literacy. They show in their paper that even though more low literate people search the internet for health information, the majority of online health information is hard to read and, therefore, inaccessible to people who are low literate. Readability is especially important for low literate people, because they struggle more with reading instructions than literate people (Mciness & Haglund, 2011).
simplified text instructions for patients who are prescribed medicines have higher compliance than the control group without the simplified prescription instructions.
Readability is also expected to be important when sending out reminders to debtors, because the reminder is expected to be readable for every debtor and the reminder should communicate the message in an apprehensive manner toward debtors in order for debtors to understand the full message.
Following from this guideline and theory, the first three hypotheses will test whether or not the usage of pictograms and simplified text in the reminder(s) actually leads to a significantly positive effect of debtors to pay off their debt to the collection agency Syncasso.
Some debtors in this research project have received two reminders during the sampling period. The fourth hypothesis will test whether the improvement from first receiving the control reminder and next receiving a simplified reminder will actually lead to more compliance among debtors. If there are differences in effectiveness for the different first and second reminders, debt collection agencies can foresee what type of reminder is the most effective to send as a first reminder and as a second reminder. There is expected that by using simpler text and supporting pictograms, debtors will understand the reminder better and will therefore pay off their debts faster.
The constructed hypotheses to answer sub-questions 1 & 2 are as following:
Hypothesis 1: The usage of pictograms in a reminder has a significantly positive effect for debtors to pay off their debts.
Hypothesis 2: The usage of simplified text in a reminder has a significantly positive effect for debtors to pay off their debts.
Hypothesis 4: Debtors who got a reminder with simplified text and pictograms as a second reminder significantly pay off more debts than debtors who receive the control letter as a second reminder.
2.3 Debt literacy
Debt literacy is an important component of financial literacy and refers to the ability to make simple decisions regarding debts. Low levels of debt literacy are common among the elderly, women, certain minorities, and people with lower incomes and wealth. It is also studied that a part of elderly think they know considerably more than they actually do (Lusardi & Tufano, 2009). Because it can be expected that people who always pay their debts are often more financially skilled (Lusardi & Tufano, 2015), it can be assumed that the debtors in this research project are overall more likely to be debt illiterate than the average Dutch population.
Additionally, it is explored that people who have high-cost borrowing fees are those who come from vulnerable demographic groups and are also less debt literate (Lusardi & Tufano, 2009), mainly because these people struggle making the right financial choices. High-cost borrowing fees arise in this research project from non-compliance and therefore it is likely to think that people in this research project who are showing non-compliance behavior after receiving the reminder(s) from the debt collection agency are less likely to be debt literate.
Furthermore, this theory can be linked to this research project by looking at the different amounts of debts and assessing the effects of simplified reminders on the different amounts of debts. Previous research also suggests that people who are debt illiterate are likely to have multiple debts or higher debts on average than debt literate people (Disney & Gathergood, 2013). People with an average higher amount of debt indicate over-indebtedness in our research which indicates low literacy, there would then be expected that simplified reminders are more effective for these people. This gives us the following hypothesis which will answer sub-question 3:
Low levels of financial literacy often indicate that these persons are low educated and are also worse in decision making and often have disadvantageous mortgages comparing to literate people (Lusardi & Tufano, 2015). Furthermore, Huston (2010) finds that the lack of financial literacy among consumers is positively associated with non-payment of debts. Courchance & Zorn (2005) show that people who have late payments often learn from bad financial events. However, low literate people seem not to learn from bad financial events due to the lack of debt literacy. In this research project there will be tested if people living on a risky address are less likely to fulfill their debt. A risky address is defined in this research project as someone who lives at an address with high risk of non-payment. This risk is defined by looking at the debt collection rate (low debt collection rate indicates risky address), or whether or not the address is of a homeless shelter or prison (homeless shelters and prisons indicate low debt collection rate, thus indicating risky address). People registered at these risky addresses are expected to have a higher chance of non-payment than people living at non-risky addresses (Courchance & Zorn, 2005). Using this information, a hypothesis is formulated to answer sub-question 4:
Hypothesis 6: Debtors living at a risky address are significantly less likely to fulfill their debt.
Furthermore, researchers analyzed the effect of age on financial literacy and the complementary consequences. Finke et al., (2016) reported that there is a consistent linear decline in average financial literacy of about one percentage point per year among respondents over 60 years old. Other research shows that interventions facilitating illiteracy among elderly include simplified communication and giving the debtor the feeling that people listen to their needs. Simplifying text and making clear that the debtor can contact the debt collection agency would be expected to contribute to compliance among elderly. Using this information, the next hypothesis is formulated which will answer sub-question 5:
2.4 Conceptual model
3. Methods
This chapter will reflect on the construction of this research project and which steps are taken to perform this research and eventually answering the research question and sub-questions. First, the research design is explained. Second, the used manipulations for the different reminders is discussed. Furthermore, the data is explained and the used analysis technique is discussed. Finally, the used variables are shown and explained.
3.1 Research design
This research project makes use of a quantitative research type. The data used in this research project is gathered from debtors who are clients from the collection agency Syncasso Nederland B.V. Syncasso sends thousands of postal reminders every month to their clients. The purpose of these postal reminders is to inform the debtor of having a debt. Furthermore, these postal reminders contain instructions in order to get rid of the debt and also contains information about consequences for the debtor of not paying their debt. The debtors in this research project are debtors who have failed to pay an outstanding fine to a Dutch public transport company.
The postal reminders which are sent out from Syncasso are manipulated in order to answer the research question and sub-questions stated earlier in chapter 1 of this thesis. The main purpose of the research is to test whether or not simplified reminders lead to increased compliance among debtors. Compliance is measured from whether or not a debtor has paid its debt.
As stated earlier, Syncasso sends thousands of postal reminders every month to their clients. This makes it possible to send tweaked postal reminders to a relatively big sample and compare the results with each other. Some clients also received a second postal reminder from Syncasso for their debt and the effects of this second postal reminder will also be analyzed in terms of compliance.
3.2 Manipulations
• Control version (appendix A). This version of the reminder is the original version of the reminder as it was been sent before this research project to debtors. This version contains relatively long sentences and words seldom used in daily conversations. • Original version with pictograms (appendix A). This version of the reminder contains
an extra page in addition to the original reminder. This extra page contains the most important information with instructions on how to pay or contact Syncasso. This version contains little text, and makes use of pictograms to make the message clearer. The usage of pictograms is suggested by DuBay (2004) and should help by better conveying the message of the reminder.
• Simplified version (appendix A). This version of the reminder is changed on the basis of suggestions from a report from Stichting Lezen & Schrijven (2017) on how to make text better readable. The changes in this version also match the suggestions given by DuBay (2004). This version contains simpler words and shorter sentences, with spaces after each paragraph.
• Simplified version with pictograms (appendix A). This version of the reminder is the same as the simplified version, but also makes use of pictograms to make the message clearer, which is also suggested by DuBay (2004).
The manipulations are randomly assigned to each debtor. From all the debtors who were 18 years or older (age at which someone is financially independent) and received the first reminder, 2596 debtors received the control version, 1838 debtors received the reminder with the simplified text, 1950 debtors received the reminder with original version with pictograms and 1889 debtors received the reminder with the simplified text and pictograms which totals 8273 sent first reminders.
For the second reminder, 2091 debtors received the control version, 1122 debtors received the reminder with the simplified text, 1195 debtors received the reminder with original version with pictograms and 1223 debtors received the reminder with the simplified text and pictograms, which totals 5631 sent second reminders.
such as: amount of debt, payment (yes = 1, no = 0), type of manipulated reminder among other variables such as sociodemographic data: age, gender, type of residence. When observing whether a debtor has paid or not, a 14-day observing period is chosen, because after the period of 14 days after receiving the reminder it cannot be concluded that the dependent variable is still a consequence of the received reminder.
Moreover, Experian PLC (a consumer reporting agency company) has classified each debtor in the used dataset into a consumer segment. These segments are created based on different variables such as demographic, financial, socio-economic, geographical, residence characteristics, lifestyle and residence-value variables. These segments made by Experian are shown in appendix B. These segments are then used to create four main classifications in the dataset in order to answer sub-question 6 which will further be discussed in chapter 3.5 of this report. The used relevant variables in the analyses are described and shown in chapter 3.5 of this report.
3.3.1 Sample
The relevant population of this research project consists of Dutch residents who are in debt. The sample also possibly includes low literate persons who do not possess of sufficient language skills. This is the case for a great part (59.6%) of debtors who are in debt at Syncasso (Keizer, 2019). There cannot be assumed that the debtors from whom the data is received is representative for low literate persons in general, because this dataset is a subset of Syncasso’s customer data file, which may possibly not be representative for low literate persons in general. It is hard to point out every low literate person out of the sample. Therefore, the reminders are sent to a big sample and thereafter, the effects of the different reminders for the low literate people are expected to be found in the whole sample. Although this is not possible from this dataset, an optimal way of looking at the effects of simplified reminders on the compliance of low literate people would be when the whole dataset consists out of low literate people.
3.4 Logistic regression
Data analysis is performed using R. To analyze the data a binary logistic regression model is utilized. The dependent variable in the analyses is whether the debtors have paid their debt or not, making it a binomial dependent variable (yes = 1, no = 0). The independent variables in the analyses are partly sociographic variables and partly behavioral variables. These independent variables are: age, gender, risky address, total sum of debt, mosaic group, type of reminder, second type of reminder. The assumptions for this type of regression will be checked in chapter 4. The assumptions for a binary logistic regression model are (Menard, 2002):
• The dependent variable is binary or dichotomous, i.e. it only contains data for the dependent variable coded as 1 (true, present, success, etc.) or 0 (false, absent, failure, etc.).
• Logistic regression requires that the observations do not come from repeated measurements or matched data.
• Independent variables should not be too highly correlated with each other.
• Logistic regression requires a large sample size. Typically, it is necessary to have a minimum of 10 cases with the least frequent outcome for each independent variable in your model.
• Logistic regression assumes linearity of independent variables and log odds. The independent variables need to be linearly related to the log odds.
3.5 Dependent variable & independent variables
In the analyses, there will be made use of one dependent variable. The first dependent variable in the dataset is payment. This variable measured whether a debtor has paid off or not the debt in 14 days after receiving the first/second reminder (Yes = 1, No = 0).
Table 1: definitions of the different dependent & independent variables used in the analyses. Variables
Dependent Variables
payment Discrete variable, defined by whether a person has paid in 14 days (1) or not (0).
Independent variables
total_debt Continuous variable, the variable can take any positive value. The value represents the total debt owed to the collection agency by a debtor.
nr_reminder Discrete variable, the variable takes 1 if the sent
reminder is the first reminder. The variable takes 2 when the sent reminder is the second reminder.
type_letter Discrete variable, the variable is defined as type of sent reminder and can take four values (Ctrl,
Ctrl&Pictograms, Simplified, Simplified&Pictograms). The variable takes Ctrl when the sent reminder is a control version of the reminder.
The variable takes Ctrl&Pictograms when the sent reminder is the original reminder with pictograms. The variable takes Simplified&Pictograms when the sent reminder is simplified and includes pictograms.
The variable takes Simplified when the sent reminder is simplified and does not include pictograms.
duration Continuous variable, defined as the time in days of which the particular file was opened by the debt collection agency.
age Continuous variable, defined as the age of a debtor in
years.
risky_address Discrete variable, defined as t (true) or f (false) for a risky address based on the amount of unpaid debts in a debtor’s zip-code.
previous_letter Discrete variable, this variable can take four values. The variable takes Ctrl when the previous sent reminder is a control version of the reminder.
The variable takes Ctrl&Pictograms when the previous sent reminder is the original reminder with pictograms. The variable takes Simplified&Pictograms when the previous sent reminder is simplified and includes pictograms.
The variable takes Simplified when the previous sent reminder is simplified and does not include pictograms. This variable is only used for people who received the second reminder.
mosaic_type Discrete variable, this variable can take 14 values. Each value represents the segment which a debtor is classified to.
These values are each explained in appendix B. mosaic_group Discrete variable, this variable can take four values
(Group 1, Group 2, Group 3, Group 4). Each value represents the group which a debtor is classified to. These groups are based on the segments from the variable mosaic_type.
Group 1 consists of people out of the segments A, B, C & D.
Group 2 consists of people out of the segments E, F & G. Group 3 consists of people out of the segments H, I, J, K & L.
4. Results
In this chapter, process of cleaning the data will be discussed. Secondly, descriptive statistics of the variables will be given along with explanation. Furthermore, output from the various binary logistic regressions will be shown and hypotheses will be answered. Finally, a summary is shown with the results of the hypotheses.
4.1 Data cleaning
At the starting point, the dataset contained 14490 observations from 8600 clients.
The dataset is cleaned in a way that the relevant variables remained and the hypotheses can be answered. The process of how this is done is explained next.
At first, all variables in the dataset are checked for strange values, such as negative values for duration, other values than 1 or 0 for the dependent variable. Next, observations with debtors under the age of 18 are been removed. People in the Netherlands under the age of 18 years old are not financially independent and can, therefore, not be used in the analyses. The sent reminders are presumably read by this group’s parents or caregivers. After removing the people under the age of 18 years old, 13904 observations from 8273 clients remain.
Furthermore, there were missings for the variables: gender, risky_address, mosaic_type and previous_letter. The missings for the variable previous_letter appeared to be all control
reminders and could then all be filled up with this value. The missings for the variables gender, age, risky_address & mosaic_type had to be imputed by using the mice package in R.
This package makes use of multiple imputation which can be used for missings which are missing at random (MAR) (Leeflang et al., 2016). MAR-type missings are missings unrelated to variables in the analysis. Missings such as for the variables gender, age, risky_address and
mosaic_type are assumed to be missing at random, because no visible patterns can be found in
the dataset. After doing the multiple imputation (MI), the dataset is now complete and has no missings anymore.
reminder is included in order to avoid matched data which is done to comply to one of the assumptions of binary logistic regression.
4.2 Descriptive statistics
In the tables below, the descriptive statistics for the relevant variables can be found for both datasets. The mean, standard deviation, minimum and maximum values are only displayed for continuous variables. The total number of observations of a certain value (0 or 1) or the total number of observations for each dummy is displayed for the discrete variables.
Table 2: Descriptive statistics of the variables for the dataset with people who got reminder #1.
Variable Obs. Mean Min. Max.
Total 0 1 Dependent variable payment 8273 6482 1791 Independent variables total_debt 8273 110.1 39.0 5253.2 type_letter 8273 Simplified/Simplified&Pictogram s/Ctrl&Pictograms/Ctrl 1838/1889/1950/2596 duration 8273 20.03 0 72 gender 8273 m/v 4130/4143 age 8273 35.24 18 101 risky_address 8273 f/t 8152/121 mosaic_group 8273 Group1/Group2/Group3/Group4 4802/1461/1184/826
Table 3: Descriptive statistics of the variables for the dataset with people who got reminder #2.
Variable Obs. Mean Min. Max.
Total 0 1
Dependent variables
type_letter 5631 Simplified/Simplified&Pictogram s/Ctrl&Pictograms/Ctrl 1122/1195/1223/2091 duration 5631 28.82 0 72 gender 5631 m/v 2902/2729 age 5631 36.03 18 98 previous_letter 5631 Simplified/Simplified&Pictogram s/Ctrl&Pictograms/Ctrl 154/173/157/5147 mosaic_group 5631 Group1/Group2/Group3/Group4 3223/986/847/575
4.3 Binary logistic regression
The binary logistic regression provides information about variables that influence the probability whether or not a debtor will pay their debt in 14 days.
Multiple models are constructed in software program R to answer the formulated hypotheses in chapter 2:
These following models are chosen to answer the hypotheses:
Model 1:
payment= α + β1*type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl + β2*age + β3*duration + β4*risky_address_t/f + β5*gender_m/f + ε
Model 1 is using the dataset only taking the first reminder into consideration.
This model is used for answering hypotheses 1, 2, 3 & 6. In this model, Ctrl (control
reminder) is the reference level for the type_letter variable and t (true) is the reference level
+ β2*age + β3*duration+
β4*previous_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl + ε
Model 2 is using the dataset only taking the second reminder into consideration. This model is used for answering hypothesis 4. Next to the same dummy variables as in model 1, this model also uses the variable previous_letter with the dummy levels: Simplified,
Simplified&Pictograms, Ctrl&Pictograms & Ctrl. These aforementioned variables are
included in the analysis, because these variables yield the highest model fit values and these variables enable the model to answer the hypotheses.
Model 3:
payment = α + β1*type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl + β2*age + β3*duration + β4*risky_address_t/f + β5*gender_m/f + β6*
type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl *total_debt + ε
Model 3 is using the dataset only taking the first reminder into consideration. This model is used for answering hypothesis 5. This model uses the same (dummy) variables as model 1, but also includes the variable gender with the dummy levels m and f. These aforementioned variables are included in the analysis, because these variables yield the highest model fit values and these variables enable the model to answer the hypotheses.
Model 4:
payment = α + β1* type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl + β2*age + β3*duration + β4*gender_m/f +
β5*age*type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl + ε
payment= α + β1*type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl + β2*age + β3*duration + β4*risky_address_t/f + β5*gender_m/f +
β6*type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl*mosaic_group_Gro up1/Group2/Group3/Group4 + ε
Model 5 is using the dataset only taking the first reminder into consideration. In addition to this, it includes a variable mosaic_group. This variable gives one of the following values
Group 1, Group 2, Group 3 & Group 4. Each value represents the mosaic group a person
belongs to.
Model 6 Mosaic groups second reminder:
payment = α + β1*type_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl + β2*age + β3*duration+
β4*previous_letter_Simplified/Simplified&Pictograms/Ctrl&Pictograms/Ctrl*mosaic_group_ Group1/Group2/Group3/Group4 + ε
Model 6 is using the dataset only taking the second reminder into consideration. In addition to this, it includes a variable mosaic_group. This variable gives one of the following values Group
1, Group 2, Group 3 & Group 4. Each value represents the mosaic group a person belongs to.
Model 5 and 6 are used for answering sub-question 6: ‘What type of reminder is the most
effective for each sub-group?’.
The sub-groups in this question refer to each mosaic group in the dataset (Group 1-4).
4.3.1 Binary logistic regression assumptions
Finally, the assumptions for a binary logistic regression models should be checked before actually running the analyses.
The assumptions for a binary logistic regression model are (Menard, 2002):
• Logistic regression requires that the observations do not come from repeated measurements or matched data.
The datasets are divided into two smaller datasets with a group which only received the first reminder and a group which received the second reminder. Each binary logistic regression model only uses data which does not come from repeated measurements or matched data.
• Independent variables should not be too highly correlated with each other.
Independent variables are not highly correlated with each other. This has been checked for in R.
• Logistic regression requires a large sample size. Typically, it is necessary to have a minimum of 10 cases with the least frequent outcome for each independent variable in your model.
Each model has at least more than 4210 observations, which is enough to satisfy this assumption.
• Logistic regression assumes linearity of independent variables and log odds. The independent variables need to be linearly related to the log odds.
This assumption can be checked with the Box-Tidwell test. This has been checked for in R.
4.3.2 Model fit
Various model fit criteria are shown in table 5-9. The model fit indicates the goodness of fit of the given model for the dataset. The first model fit criterion is the Pseudo R2 from McFadden. This R2 is a number between 0 and 1, which indicates a high model fit when it is close to 1 and indicates a low model fit when it is close to 0. Lower than 0.1 usually indicates a bad model fit, between 0.1 and 0.2 usually means a good fit. Above 0.2 means the model has an excellent fit.
Adding a variable should lower the AIC and thus add statistical power to the model. This explains why the models in chapter 4.3 differ between chosen variables.
4.4 Results hypotheses
4.4.1 The effect of simplified text and usage of pictograms in the first debt reminder
The constructed hypotheses to answer sub-question 1 are as following:
‘Hypothesis 1: The usage of pictograms in a reminder has a significantly positive effect for debtors to pay off their debts.’
‘Hypothesis 2: The usage of simplified text in a reminder has a significantly positive effect for debtors to pay off their debts.’
‘Hypothesis 3: The usage of simplified text and pictograms together in a reminder have a significantly positive effect for debtors to pay off their debts.’
The output for model 1 together with the model fit values are shown below in table 4. It is a binary logistic regression of model 1 with payment as a dependent variable and type_letter,
age, duration, risky_address, gender as independent variables.
Table 4: binomial logit model 1 Model 1
Variable Coefficient Std. error
intercept 0.676*** 0.119 type_letterSimplified -0.305*** 0.092 type_letterSimplified&Pictograms -0.212* 0.091 type_letterCtrl&Pictograms -0.358*** 0.091 age -0.009*** 0.003 duration -0.093*** 0.004 risky_addresst -2.081*** 0.598 genderf -0.052 0.067 Observations 6204 Pseudo R2 (McFadden) 0.162 Hit-rate 81.6% AIC 5497.2
***, **, * and . consecutively stand for the 0.001, 0.01, 0.05 and 0.1 significance level
effect for debtors to pay off their debts. Hypotheses 1, 2 & 3 can be rejected on the basis of model 1.
4.4.2 The effect of simplified text and usage of pictograms in the second debt reminder
The constructed hypothesis to answer sub-question 2 is as following:
‘Hypothesis 4: Debtors who got a reminder with simplified text and pictograms as a second reminder significantly pay off more debts than debtors who receive the control letter as a second reminder.’
The R-output for model 2 together with the model fit values are shown below in table 5. It is a binary logistic regression of model 2 with payment as a dependent variable and type_letter,
age, duration & previous_letter are used as independent variables.
Table 5: binomial logit model 2 Model 2
Variable Coefficient Std. error
intercept 0.822*** 0.162 type_letterSimplified 0.053 0.130 type_letterSimplified&Pictograms -0.047 0.129 type_letterCtrl&Pictograms -0.122 0.130 age -0.020*** 0.004 duration -0.087*** 0.004 previous_letterSimplified 1.085** 0.333 previous_letterSimplified&Pictograms 1.594*** 0.300 previous_letterCtrl&Pictograms 1.093*** 0.320 Observations 4223 Pseudo R2 (McFadden) 0.205 Hit-rate 82.3% AIC 2991.2
***, **, * and . consecutively stand for the 0.001, 0.01, 0.05 and 0.1 significance level
As can be seen in the output for model 2, there is no significant different effect for the type of sent reminders comparing to the control reminder. Therefore, on the basis of model 2, there can be stated that hypothesis 4 is rejected.
4.4.3 The effect of amount of a debt on the effectiveness simplified reminders
The R-output for model 3 together with the model fit values are shown below. It is a binary logistic regression of model 3 with payment as a dependent variable and type_letter, gender,
total_debt, age, duration & risky_address as independent variables with an interaction effect
between type_letter and total_debt. Table 6: binomial logit model 3
Model 3
Variable Coefficient Std. error
intercept 0.603*** 0.129 type_letterSimplified -0.161 0.138 type_letterSimplified&Pictograms -0.025 0.154 type_letterCtrl&Pictograms -0.309. 0.161 Gender -0.053 0.067 total_debt 0.001 0.001 age -0.008*** 0.003 duration -0.094*** 0.004 risky_addresst -2.076*** 0.598 type_letterSimplified:total_debt -0.001 0.001 type_letterSimplified&Pictograms:total_debt -0.002 0.001 type_letterCtrl&Pictograms:total_debt -0.000 0.001 Observations 6204 Pseudo R2 (McFadden) 0.162 Hit-rate 81.5% AIC 5501.7
***, **, * and . consecutively stand for the 0.001, 0.01, 0.05 and 0.1 significance level
As can be seen in the output for model 3, there is no significant effect on the interaction between
total_debt and type_letter. There can be concluded that the amount of a debt does not matter
for the effect of a simplified text or usage of pictograms on paying off debts. On the basis of model 3, there can be stated that hypothesis 5 is rejected.
4.4.4 The effect of a debtor living on a risky address on paying a debt.
The constructed hypothesis for answering sub-question 4 is as following:
‘Hypothesis 6: Debtors living at a risky address are significantly less likely to fulfill their debt.’
As can be seen in the output for model 1 in table 4, there is a significant positive effect of
riskadrest on the dependent variable payment. This means that debtors living on a risky
4.4.5 The effect of age on the effectiveness of simplified reminders
The constructed hypothesis for answering sub-question 5 is as following:
‘Hypothesis 7: Debtors who are relatively older pay off their debts faster with the simplified reminders than debtors who are relatively young.’
The R-output for model 4 together with the model fit values are shown below. It is a binary logistic regression of model 4 with payment as a dependent variable and type_letter, gender,
age*type_letter, age, duration & risky_address as independent variables.
Table 7: binomial logit model 4 Model 4
Variable Coefficient Std. error
intercept 0.668*** 0.177 type_letterSimplified -0.276 0.258 type_letterSimplified&Pictograms -0.024 0.257 type_letterCtrl&Pictograms -0.543* 0.255 genderf -0.052 0.067 age -0.008. 0.005 duration -0.094*** 0.004 risky_addresst -2.085*** 0.598 type_letterSimplified:age -0.001 0.007 type_letterSimplified&Pictograms:age -0.005 0.007 type_letterCtrl&Pictograms:age 0.005 0.007 Observations 6204 Pseudo R2 (McFadden) 0.162 Hit-rate 81.6% AIC 5501.0
***, **, * and . consecutively stand for the 0.001, 0.01, 0.05 and 0.1 significance level
Hypothesis 7 suggests that there is an interaction effect between age and type of reminder and that this effect is significantly positive.
Model 4 shows that there is no significant interaction effect between age and type_letter. Therefore, there can be concluded that that hypothesis 7 is rejected on the basis of model 4.
is the most effective for each sub-group?’. The outputs of these two tables will further be
discussed in chapter 5 of this report.
Table 8: binomial logit model 5 Model 5
Variable Coefficient Std. error
intercept 0.453*** 0.129 type_letterSimplified -0.125 0.120 type_letterSimplified&Pictograms -0.224. 0.125 type_letterCtrl&Pictograms -0.243. 0.125 mosaic_groupGroup2 0.427** 0.161 mosaic_groupgroup3 0.464** 0.171 mosaic_groupgroup4 0.254 0.210 age -0.010*** 0.003 duration -0.091*** 0.004 risky_addresst -1.591*** 0.476 genderf 0.003 0.067 type_letterSimplified:mosaic_groupGroup2 -0.069 0.245 type_letterSimplified&Pictograms:mosaic_groupGroup2 -0.147 0.245 type_letterCtrl&Pictograms:mosaic_groupGroup2 -0.284 0.247 type_letterSimplified:mosaic_groupGroup3 0.047 0.253 type_letterSimplified&Pictograms:mosaic_groupGroup3 0.137 0.259 type_letterCtrl&Pictograms:mosaic_groupGroup3 -0.058 0.257 type_letterSimplified:mosaic_groupGroup4 -0.464 0.339 type_letterSimplified&Pictograms:mosaic_groupGroup4 -0.246 0.325 type_letterCtrl&Pictograms:mosaic_groupGroup4 -0.211 0.330 Observations 6218 Pseudo R2 (McFadden) 0.161 Hit-rate 80.5% AIC 5500.7
***, **, * and . consecutively stand for the 0.001, 0.01, 0.05 and 0.1 significance level
Table 9: binomial logit model 6 Model 6
Variable Coefficient Std. error
mosaic_groupgroup 2:type_letterSimplified 0.269 0.342 mosaic_groupgroup 3:type_letterSimplified 0.232 0.334 mosaic_groupgroup 4:type_letterSimplified 0.545 0.448 mosaic_groupgroup 2:type_letterSimplified&Pictograms 0.744* 0.326 mosaic_groupgroup 3:type_letterSimplified&Pictograms 0.066 0.350 mosaic_groupgroup 4:type_letterSimplified&Pictograms 0.757. 0.426 mosaic_groupgroup 2:type_letterCtrl&Pictograms 0.644* 0.320 mosaic_groupgroup 3:type_letterCtrl&Pictograms -0.081 0.337 mosaic_groupgroup 4:type_letterCtrl&Pictograms -0.101 0.461 Observations 4210 Pseudo R2 (McFadden) 0.184 Hit-rate 83.3% AIC 3058.64
4.5 Summary hypotheses and results
Hypothesis Description Result
H1 Hypothesis 1: The usage of pictograms in a reminder has a
significantly positive effect for debtors to pay off their debts.
Rejected
H2 Hypothesis 2: The usage of simplified text in a reminder has a
significantly positive effect for debtors to pay off their debts.
Rejected
H3 Hypothesis 3: The usage of simplified text and pictograms
together in a reminder have a significantly positive effect for debtors to pay off their debts.
Rejected
H4 Hypothesis 4: Debtors who got a reminder with simplified text
and pictograms as a second reminder significantly pay off more debts than debtors who receive the control letter as a second reminder.
Rejected
H5 Hypothesis 5: Simplified reminders are more effective for
higher amounts of debts than for lower amounts of debts.
Rejected
H6 Hypothesis 6: Debtors living at a risky address are significantly
less likely to fulfill their debt.
Supported
H7 Hypothesis 7: Debtors who are relatively older pay off their
debts faster with the simplified reminders than debtors who are relatively young.
Rejected
5. Conclusions
In this chapter, the results of the hypotheses will be visualized by using the conceptual model again. Furthermore, conclusions of the results will be shown. Lastly, the limitations of this research project will be discussed.
5.1 Conceptual model results
In the figure below, the results are visualized in the conceptual model.
5.2 Main findings
The purpose of this research project is to assess the effects of tweaks in the communication toward debtors living in the Netherlands and assessing the effects hereof. The main communication toward debtors is done using one page long letters reminding debtors to pay their debt off. The main research question of this research project is:
What type of debt reminder is the most effective for debtors?
The results show that, in general, the usage of simplified text and pictograms in reminders are ineffective comparing to the original reminder.
5.2.1 Answering sub-questions 5.2.1.1 Sub-question 1
The first sub-question is as following:
Is the usage of fitting pictograms and a simplified text an effective way in order to make the debtors pay off their debt?
three hypotheses are constructed to answer this sub-question based on research from Mayeaux et al. (1996), Andrus & Roth (2002) and Vetter et al. (2014). These research articles show that using simplified instructions (verbally or written) are effective in terms of compliance.
The results in this research project show that, in general, the usage of simplified text and pictograms when sending the first debt reminder is ineffective. Also, when looking at the effects of the usage of simplified text and pictograms when sending the second debt reminder seem to be ineffective.
However, when dividing the debtors into mosaic groups based on Experian’s consumer segments, the usage of simplified text and pictograms and the usage of the original reminder with pictograms seem to be significantly positively influencing the probability of payment for
5.2.1.2 Sub-question 2
The second sub-question is:
For the group of debtors who received two reminders, which second reminder was the most effective?
For answering this sub-question, one hypothesis is constructed. This hypothesis is built on the research studies of Mayeaux et al. (1996). They show that people who first received normal instructions for health prescriptions and thereafter receiving simpler health instructions showed greater health compliance than people who did not receive simpler health instructions. The results from this thesis show that there is no significant effect of tweaked reminders for the second reminders in terms of payment.
5.2.1.3 Sub-question 3
The third sub-question is:
Does the effectiveness of a reminder change between different amounts of debts?
For answering the third sub-question, a hypothesis is constructed based on research from Lusardi & Tufano (2009). They discuss in their paper that people who are used to build up high debts are more likely to be debt low literate. The results from this thesis show that there is no significant effect of the amount of a debt on effectiveness of simplified reminders and the usage of pictograms.
5.2.1.4 Sub-question 4
The fourth sub-question is as following:
How does the living situation affect the effectiveness of each reminder?
5.2.1.5 Sub-question 5
The fifth sub-question is as following:
How does a debtor’s age affect the effectiveness of each reminder?
Finke et al. (2016) reported in their research paper that there is a consistent linear decline in average financial literacy of roughly one percentage point per year among respondents over 60 years old. The hypothesis constructed from this scientific paper tests whether or not the simplified reminders are more effective for older debtors than for older debtors. The results show that debtors who are relatively older do not have a significantly higher probability of paying off their debt with simplified reminders than debtors who are relatively young.
5.2.1.6 Sub-question 6
The last sub-question is as following:
What type of reminder is the most effective for each sub-group?
All the debtors have been assigned to mosaic groups by Experian’s consumer segmentation. After assigning every debtor to a mosaic group, it can be seen that Group 2 responds significantly positive (in terms of payment) to the usage of pictograms and simplified text together and to the original reminder with the usage of pictograms.
This group can be characterized as a middle-class group with an average income, average age and low number of children.
5.3 Discussion findings
6. Recommendations
In this chapter, for each target audience, the recommendations will be discussed. Finally, suggestions for future research will be discussed.
6.1 Recommendations for debt collection agencies
Based on this research project, debt collection agencies should classify each debtor into a mosaic group based on different variables. Following from the results from this research project, the middle-class group should be targeted with simplified reminders (simplified text and usage of fitting pictograms) in order to increase compliance. The other groups can still be targeted with the original reminders as it seems to be ineffective to target these groups with simpler reminders. Although, this research project suggests these previous recommendations, it would be better to do this research again for debtors who have debts at different companies and look at the effects simplified reminders have for these people. It may be possible that debtors having a debt at other companies such as health insurance companies and energy companies react differently to simplified reminders (partly) caused by different compositions of literate/low literate people in these samples. Debt collection agencies should classify debtors with care into a group based on relevant variables and should then target the various groups appropriately. Companies must take the benefits of targeting people effectively into consideration.
Furthermore, debtors who have multiple debts a year at a debt collection agency could be asked to fill in a test to measure its literacy levels. Low literate people should thereafter be targeted with simplified reminders with supporting pictograms in order to increase compliance (Mayeaux et al., 1996).
6.2 Recommendations for policy makers
consumers of the dangers of over-indebtedness and should stimulate people to be cautious with their expenditure.
Low literate people often feel ashamed by their financial status, low literate people should be able to speak with someone about its finances causing the low literate person to feel less ashamed. A safe environment should give the low literate person a safe feeling, causing the person to openly talk about his/her literacy levels. The low literate person should get to know that it is very normal to develop itself during his/her whole lifetime, which will hopefully eventually help by trying to develop low literate persons causing them to become more literate. Policy makers should let municipalities screen people who are looking for a job, or people who are signing up for benefits, because these groups consist of more low literate people on average. After selecting low literate people out of these groups, social services should help low literate people by offering courses to improve their literacy. Courses can consist of language courses for the Dutch language, or more specific language courses for company related jargon which can be helpful by working in a specific sector.
6.3 Limitations
During the process of this thesis, various limitations came to mind. First of all, in this research project, there is made use of letters sent with postal services only. Communication between people and companies becomes more and more digital nowadays. It is possible that people tend to ignore postal letters, because they are out of date. Debt collection agencies also make more and more use of e-mails and digital communication forms. Effects of simplified e-mails should also be taken into consideration in future research.
7. References
Andrus, M. R., & Roth, M. T. (2002). Health literacy: a review. Pharmacotherapy: The
Journal of Human Pharmacology and Drug Therapy, 22(3), 282-302.
Baker, D. W., Gazmararian, J. A., Sudano, J., & Patterson, M. (2000). The association between age and health literacy among elderly persons. The Journals of Gerontology Series
B: Psychological Sciences and Social Sciences, 55(6), S368-S374.
Christoffels, I., Baay, P., Bijlsma I., & Levels, M. (2016). Over de relatie tussen
laaggeletterdheid en armoede. ‘s Hertogenbosch: ecbo, Amsterdam: Stichting Lezen en
Schrijven.
Courchane, M., & Zorn, P. (2005). Consumer literacy and credit worthiness. Proceedings,
Federal Reserve Bank of Chicago.
Disney, R., & Gathergood, J. (2011). Financial literacy and indebtedness: new evidence for UK consumers. The University of Nottingham, 11-05.
Disney, R., & Gathergood, J. (2013). Financial literacy and consumer credit portfolios. Journal of Banking & Finance, 37(7), 2246-2254.
DuBay, W. H. (2004). The Principles of Readability. Online Submission.
Finke, M. S., Howe, J. S., & Huston, S. J. (2016). Old age and the decline in financial literacy. Management Science, 63(1), 213-230.
Huston, S. J. (2010). Measuring financial literacy. Journal of Consumer Affairs, 44(2), 296-316. Illiteracy. (2019). In Cambridge Dictionary (2019). Cambridge, United Kingdom: Cambridge University Press
Keizer, M., (2018b). Lezen ≠ Begrijpen: De invloed van beperkte leesvaardigheid op de
omgang met financiële problemen. Eindrapport van het onderzoeksproject Lezen ≠ Begrijpen.
Online geraadpleegd op www.lezenisnietbegrijpen.nl
Kirsch, I. S. (1992). Beyond the School Doors: The Literacy Needs of Job Seekers Served by
the US Department of Labor.
Kirsch, I. S. (1993). Adult literacy in America: A first look at the results of the National Adult
Literacy Survey. US Government Printing Office, Superintendent of Documents, Washington,
DC 20402 (Stock No. 065-000-00588-3)
Leeflang, P., Wieringa, J. E., Bijmolt, T. H., & Pauwels, K. H. (2016). Modeling markets. Springer-Verlag New York.
Lusardi, A., & Tufano, P. (2009). Teach workers about the perils of debt. Harvard Business
Review, 87(11), 22-24.
Lusardi, A., & Tufano, P. (2015). Debt literacy, financial experiences, and overindebtedness. Journal of Pension Economics & Finance, 14(4), 332-368.
Mayeaux, J. E., Murphy, P. W., Arnold, C., Davis, T. C., Jackson, R. H., & Sentell, T. (1996). Improving patient education for patients with low literacy skills. American family
physician, 53(1), 205-211.
Mcinnes, N., & Haglund, B. J. (2011). Readability of online health information: implications for health literacy. Informatics for health and social care, 36(4), 173-189.
Menard, S. (2002). Applied logistic regression analysis (Vol. 106). Sage. OECD. (2005). OECD Annual report 2005. Geraadpleegd van
https://www.oecd.org/about/34711139.pdf
Plimpton, S., & Root, J. (1994). Materials and strategies that work in low literacy health communication. Public Health Reports, 109(1), 86.
PwC. (2018). Maatschappelijke kosten laaggeletterdheid. Visited at
https://www.lezenenschrijven.nl/uploads/editor/PwC_-_Rapport_maatschappelijke_kosten_laaggeletterdheid_-_April_2018_def.pdf
Rijksoverheid, Van Putten, B., & Schoot Uiterkamp, T. (2017). SCHULDHULPVERLENING
IN NEDERLAND. Geraadpleegd van
https://www.rijksoverheid.nl/binaries/rijksoverheid/documenten/rapporten/2017/04/24/eindra
pportage-schuldhulpverlening-in-nederland/eindrapportage-schuldhulpverlening-in-nederland.pdf
Ratzan S.C., Parker R. (2000). Introduction. In: C.R. Selden, M. Zorn, S.C. Ratzan, R.M. Parker (Eds.), National Academies of Medicine Current Bibliographies in Medicine: Health Literacy. Bethesda, MD: National Institutes of Health, U.S. Department of Health and Human Services, 2000.
Stichting Lezen & Schrijven. (2012). Feiten & Cijfers. Visited at https://www.lezenenschrijven.nl/over-laaggeletterdheid/feiten-cijfers/ Stichting Lezen & Schrijven. (2016). Feiten & Cijfers. Visited at https://www.lezenenschrijven.nl/over-laaggeletterdheid/feiten-cijfers/
Vetter, T. R., Downing, M. E., Vanlandingham, S. C., Noles, K. M., & Boudreaux, A. M. (2014). Predictors of patient medication compliance on the day of surgery and the effects of providing patients with standardized yet simplified medication instructions. Anesthesiology:
8. Appendices
Appendix A: The various reminders for the first and second reminder
Appendix C: R-output Binomial logistic regression models
Model 1:
Model 2:
Model 4:
