The effect of linguistic style on the selling time of real estate properties
Rob Sijs, 11880384
Faculty of Economics and Business
Bachelor’s Thesis
Frederik Situmeang
2 Statement of Originality
This document is written by Student Rob Sijs who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Abstract
House descriptions are major sources of information which you write when selling your house using Funda, but which linguistic style should you use to have the fastest selling time? To explain the effect of linguistic styles on selling times, this study uses the signalling theory. The signalling theory explains that people make decisions on the basis of different signals, linguistic styles can be seen as signals which affect the selling time of a house. The study hypothesizes that if the ratio of positive affective language in house descriptions is increased, that the selling time decreases and that if the ratio of risk language in house descriptions increases, that selling time increases. The hypotheses are tested with data of all the houses in the Netherlands which were sold via Funda in 2018. The analysis supported the hypothesis which indicates that positive emotional language reduces selling times, but there was no clear result on the effect of risk language on the selling times, this should be further researched.
4
Table of contents
Introduction...5 Theoretical framework...8 Signalling theory ...8 Linguistic style...9 Methodology ... 15 Dataset ... 15Linguistic inquiry and word count (LIWC) – analysing content and linguistic style of a text ... 16
Model ... 17 Dependent variable ... 17 Independent variables... 17 Control variables ... 18 Multicollinearity ... 19 Results ... 20 Discussion ... 25 Conclusion ... 27 Bibliography:... 28 Appendix ... 31 Appendix 1 ... 31
5
Introduction
Funda.nl has become the most frequently used website in the Netherlands for people to explore houses when planning on buying a new house in the 21st century. Funda.nl is a website which was created in 2001 by the Nederlandse Vereniging van Makelaars (NVM), which is the biggest real estate association in the country (Over Funda, n.d.). The website has 43 million visitors a month, a spontaneous brand awareness of 93%, and has 5000 real estate agents connected to the website which makes it the largest real estate website in the Netherlands (Over Funda, n.d.).
The pages on Funda of each listed house include data including pictures, asking price, location and a description about the property which is usually written by a real estate agent. These property descriptions will be the data focused on in this research. Only the descriptions are used as data for this research because this research focuses on the effect of linguistic properties on selling times. Descriptions are created for the listed house, the seller can easily change the way it is written and can think about which style to use, that is why this research only analyses this effect. Properties of the house as for instance if the house is detached or if it is an apartment cannot be (easily) changed by the seller of the house, this shaped the focus of this research.
Text is a really important way of communicating for humans. Written texts are built of two main components, which are the contents of the text and the linguistic style of the text. Previous research concluded that humans are highly attentive to the conveyance of messages (Ludwig et al., 2013). Tausczik and Pennebaker (2010, P. 25) wrote: “Language is the most common and reliable way for people to translate their internal thoughts and emotions into a form that others can
understand. …. They are the medium by which cognitive, personality, clinical, and social psychologists attempt to understand human beings”. Recent research showed that online reviews were able to influence people’s choices by using certain kinds of content and linguistic style (Ludwig et al., 2013). While this has been researched for online reviews and the effect of content and linguistic style on conversion rates, this effect has not been researched for the selling times of real estate objects on
6 Funda.nl. This research will look at the effect of linguistic style and content on the selling time of properties on Funda.nl.
Previous research has focused on linguistic style and content in descriptions of crowdfunding projects, this study took the outcomes of these studies as a starting point to discover the effects of descriptions on selling times on Funda.nl. Recent research showed that funding outcomes in civic crowdfunding can be improved by the use of psychological language dimensions, but can also be worsened by the use of extensive use of social language (Lee et al., 2019). Another study looked at reward-based crowdfunding and how entrepreneurs’ use of narratives may cause for a change of attitude in potential funders (Allison et al., 2017). These previous studies are accounting for multiple implications that descriptions may change the attitude towards certain investments by using
different kinds of linguistic styles and contents. This study will try to gain further insights on this effect, but in another industry: the real estate industry. One way to find these further insights is by using Linguistic inquiry and word count (LIWC) to analyse the data and run regressions on the analysed data.
The reason why outcomes from the previous studies on crowdfunding performance cannot be extended to the real estate industry is because the industries differ too much from each other. It might be that the effects in both industries are the same, but this cannot be assumed without further research. The crowdfunding industry is an industry in which you invest in projects where the
outcome is mostly unknown: the project might, or might not be successful. In the real estate industry, it is known what you buy, you buy a house in which you had a viewing which means you know what to expect. The crowdfunding industry therefore is more speculative than the real estate industry. This could affect the impact of descriptions as you are more dependable on descriptions with crowdfunding projects than with real estate properties. This research will try to gain insights on this effect.
7 Another variable that will be used in this research is the selling price of the properties. One of the main filters which people can use to search on Funda.nl, is the price range. People go to their banks and check what they are able to get as a mortgage. The height of their mortgage is connected to their incomes and therefore welfare, this mortgage decides in what price range someone can search for a house. This study will use selling prices of the properties to see if this price has a moderating role on the strength of the effect of the description on selling time. Selling price is used as a moderator because the expectations are that people who are searching in different price ranges, might have different searching patterns. The previous literature of Sjostrom et al. (2016) indicates that a high price together with a high quality product make people perceive the product as a luxury product. Sjostrom et al. (2016) also write that people with higher incomes buy more luxury products than middle-class people. This indicates that the signal of a higher price might indicate that the house is a luxury product which can change the selling times. If a house is sold for 140.000 euro’s, it might be the case that this is a starter which buys his or her first house, their requirements and the way that they are searching are expected to be different from somebody who searches for a house in a price range that is above one million euros. For somebody who buys a relatively cheap house it might be perceived as a generic product, while somebody who buys a more expensive house it might be perceived as a luxury product. The outcomes of this study might provide people, and especially real estate agents, with useful insights for writing a description for a house that they want to sell.
The literature gaps which this research tries to cover is what the effects of linguistic style and content of real estate descriptions are on selling times of a house. This research also tries to answer if selling prices have a moderating effect on these selling times. For this the main question of this research is: ‘Does the linguistic style of a housing description affect the selling time of a house which is for sale?’. The outcome of this study can be used by real estate agents in writing descriptions for properties. Writing optimal descriptions is important for real estate agents because they are in most cases paid on the basis of a broker fee (Dammingh, 2018). The broker fee is in most cases a
8 the seller of the property. Creating an optimal description which shortens the selling time of a house means that a real estate agent earns his broker fee in a shorter time. Also, if sellers are willing to sell their house as quick as possible due to for instance a divorce or due to the fact that they already bought a new house, closing the literature gap might help them in achieving the fastest selling time possible.
Theoretical framework
Signalling theory
Once someone starts searching for a house, the searching person comes by different properties with different descriptions and information. While searching for a new house there is information asymmetry. The sender of the information, the ones that sell the property, send out information to the receiver, the possible buyer of the house. This information is sent out using signals and the theory behind it is called the signalling theory (Connelly et al., 2011).
The signalling theory was introduced by Michael Spence in his essay ‘Job market signaling’ (1973). Spence wrote that for employers to find employees, they are having an information asymmetry, the applicant knows about what he is capable of but needs to send this information to the employers in order to stand out. Spence focused on education level as the information that applicants can send to the employer to signal their capabilities to them. The employer will make a choice on who he or she hires on the basis of the signals that is received.
The signalling theory can be used to describe the behaviour when two parties have access to different information, in this case the seller and the (potential)buyer of a real estate property are the two parties. As Connelly et al. (2011) state in their research: ‘Information affects the decision-making processes used by individuals in households, businesses, and governments.’. This is why the seller needs to think of what information he sends to potential buyers of the property. The information is send using signals. For instance: ‘The house has been repainted last month.’, this is a positive signal,
9 and: ‘The house has mise in the basement.’ is a negative signal. These signals will most likely have an effect as they provide the receiver with information which will affects its decision-making processes. The signals which are emitted by the description pages of the properties are send using texts and numbers. This research looks at specific signals, in the form of linguistic styles, and find out what their effects are on the selling times of properties.
Linguistic style
The main signal that this research is focussing on is the signal of linguistic style. The linguistic style can occur in many different forms. A linguistic style can for instance be negative, this means that a high ratio of negative words is used in the description. The signal that this negative linguistic style could give to a potential buyer is that the quality of the house is low. As written in the
introduction, Funda communicates information in various forms like pictures and text. As written texts have different linguistic styles based on the ratio of linguistic categories in the descriptions, the signals that are given by the linguistic style might influence the selling time. The signal of every hypothesis in this research will be established before the formulation of each hypothesis.
Previous studies showed that humans use the linguistic style of a written review in order to make their decisions about the product that the review is about. It showed that various interest groups have different linguistic styles (Ludwig et al., 2013). If the linguistic style of a review matches the linguistic style of the products’ interest group, then this is called a linguistic style match (LSM). Ludwig et al. (2013) found that a positive change in this linguistic style match between a products’ interest group and the product review had a positive effect on the conversion rates of this product. A similar effect has been found by Allison et al. (2017). Alison et al. (2017) found that when a seeker for crowdfunding investments is able to adopt a group identity of his prospective funders in his
crowdfunding project, that this has a positive relationship on his crowdfunding performance. So, by talking in the same style as the prospective funders, crowdfunding ventures are able to boost their crowdfunding performance. This research will not look for a possible linguistic style match of the descriptions with the target group of a certain property, but this literature proofs that there is an
10 effect of linguistic style on outcomes within industries other than the real estate industry and implies that this could also be the case in the real estate industry.
Another study found that in civic crowdfunding, which is crowdfunding for public goods and services, positive affective language had a positive effect on the funding success of the crowdfunding project (Lee et al., 2019). This study also found that the use of risk language, language which disclose risk-related information, has a negative effect on the funding success of a civic crowdfunding project.
Juanchich et al. (2012) conducted research on how positive and negative linguistic styles effected the risk perception of the participants. The participants of the study got to read investment scenarios in which a friend communicated the possible occurrence of both a bad or a good outcome. Afterwards the researchers conducted the risk perception of these investment scenarios and found out that if someone communicated a good outcome that the risk was perceived different then when someone communicated a bad outcome. With a good outcome, 51.5% of the people took the positive possibility that is given as being a likelihood-communication device. This means that if a positive advice is given on an investment, that 51,5% takes this as a sign that it is likely to get a positive outcome, while with a negative possibility communicated, 38,5% of the participants took the possibility as being a likelihood-communication device (Juanchich et al., 2012). This means that the message which is send, the possibility, is perceived different when the content differs from positive or negative.
The signal which is send by positive affective language in the descriptions is that there is a positive signal send through the information to the receiver through text. The positive affective language informs the receiver about the likelihood of ‘profit’ and the likelihood of a good outcome when buying this particular house of which the description is from.
On the basis of the literature described above, this research hypothesizes the following:
11 To add to the previous hypothesis, this research will also look at possible detrimental effects of over-emoting in descriptions. Stephens et al. (2019) showed in their research that when
spokespersons need to apologize for something in a speech that using various levels of emotions have different effects on their credibility. If the spokesperson emote a low degree of sadness combined with a large amount of words that indicate sincerity, that this was perceived with a high degree of credibility. But, if the degree of sadness increases, the credibility decreases, this is an example of when over-emoting diminishes the positive effect of the high amount of sincere words (Stephens et al., 2019).
Also, Parhankangas and Ehrlich (2014) showed in their study that for an entrepreneur who seeks an investment from business angels, only moderate amounts of positive emotions in their language use leads to increasing chances of receiving an investment or receiving an invitation to present their business plan in person. The positive relationship was only present up to a certain point. After going over a certain amount of positive emotions, the effect diminished. Exorbitant amounts of positive language damages the attributes of sincerity, likability, and predictability. This effect was found for angel investors, but cannot be extended to the real estate market yet, still it is plausible that the effect is the same for the real estate market. Therefor this study expects that showing exorbitant levels of positive emotions, will diminish the positive effect of positive affective language used in real estate object descriptions on selling times.
The signal that can be send from the sender to the information receiver when using over-positive over-positive affective language in the descriptions is that you are not credible. As mentioned above the over emoting can diminish your credibility, which means that the receiver does not believe that the real image is as positive as the information shows him or her. Therefor:
H2: Over-positive affective language used in real estate object descriptions has detrimental effects of
12 As already mentioned earlier in this research, risk language can also influence the perception of the receiver of the information.
Lee et al. (2019) found that risk language has a negative effect on civic crowdfunding success and outcome. The findings of Lee et al. (2019) are supported by Allison et al. (2015) who writes that a greater degree of risk-taking language results in an increase in time needed to fund a microloan. Literature highlights that risk and rewards are two essential components of an investment proposal as they define the amount of risk that is anticipated by the entrepreneur and the amount of profit expected from the investment (Macmillan et al., 1985). In the same way, expressing words that signal risks can be a reason for longer selling time.
The signal that is send by risk language in the information from the sender to the receiver is that it warns for possible losses. As Macmillan et al. (1985) stated, risk and rewards are key
components of the profit expected from an investment. When more information about risk is send to the receiver, then this decreases the expectancy of profit from an investment, therefor it can also decrease the attractiveness of the investment, in this case a house, and increase selling time. This creates the following hypothesis:
H3: Risk language in real estate descriptions increase selling time.
The follow up hypothesis for the third hypothesis is related to degrees of risk and the effect on selling time moderated by selling price. Lee et al. (2019) wrote how an increasing rate of risk related information has a negative effect on crowdfunding outcomes. What the research did not specify was if this effect was also stronger or weaker for different price ranges and for different levels of investors’ wealthiness. Riley & Chow (1992) did a research on risk aversity and the effect of wealth on risk aversity. They measured relative risk aversion by taking the ratio of risky assets to wealth. Riley & Chow (1992) found in their research on risk aversity that risk aversity decreases as wealth increases. They argue that people with a low income do not have flexibility in their budgets, their income goes straight to their fixed charges. The research writes: ‘The ability to take on more risk can
13 be expected to translate into the willingness to accept more risk, because the consequences of a bad investment decision are less severe if one has income beyond some subsistence level.’ (Riley & Chow, 1992)
Kumar et al. (2015) also studied what the effect of wealth is on risk tolerance for investors. Even though the research was conducted in Pakistan, which is a non-western country, they found the same results as Riley & Chow (1992) but then for investors: wealthier investors are more risk tolerant than less wealthy investors. Kumar et al. (2015) used questionnaires in order to get to their findings. These questionnaires were developed by Dow Jones and Company in 1998. Kumar et al. (2015) argue that the reason that wealthier people are more risk tolerant than less wealthy people is because losses in investments of wealthier investors do not distress because they would not affect their standard of living.
The assumption can be made that the most expensive houses are bought by the wealthiest customers while the properties in the lowest price range are bought by the less wealthy customers. Therefore, this research expects that the negative effect of risk related words in housing descriptions become less strong as property prices become higher.
The signal send is the same as for the previous hypothesis, but then it might be moderated by the house price. The hypothesis is based on that a negative signal might be diminished in its effect once the house prices are higher. This is because the searchers of more expensive houses might be less vulnerable for being affected by a negative signal as they are less risk sensitive. The fourth hypothesis therefore is:
H4: The strength of the negative effect of risk language in real estate descriptions on selling times
become less strong as housing prices rise.
However, there is another factor which could be more relevant to the effect of risk language on selling times which is market heat. Clayton et al. (2008) wrote in their investigation on liquidity in the private real estate market that when the prices in the real estate markets are rising, that sales
14 activity increases and average time-on-market decreases for a property to be sold. The contrary of this effect holds as well: when prices fall the markets typically exhibit a decrease in sales and an increase in time-on-market (Clayton et al., 2008). The traditional explanation given for this phenomenon in which sales activity decrease when house prices decrease, is that sellers are ‘irrationally’ and refuse to recognize declines in the value of their property therefor continue their house at a higher price. For hot real estate markets the time-on-market falls because buyers extrapolate the recent price movements into the future. The buyers think that the prices will continue to go up so their house price expectation is irrational (Clayton et al., 2008).
Krainer (2001) indicates that the heat of the market can be measured by house prices. When prices rise, the market is warming up, while when prices fall the market is getting colder. Novy-Marx (2009) also found that when the real estate market is heated, that there are more buyers than sellers. This makes the average selling time relatively low. In a cold real estate market there are more buyers than sellers which results in a longer selling time (Novy-Marx, 2009). The way to check for the heat of the real estate market is by looking at the price indexes of the real estate market. Because the outcome of H4 might be moderated by the heat of the market, this study adds a hypothesis to check for this effect.
For the upcoming hypothesis the signal of risk in housing descriptions might be less
influential when the heat of the market rises. Because the scarcity of houses is higher, people might be less affected by risk language as they might be more risk tolerant because houses are scarce. One can think of a situation where houses are sold very fast which makes it hard to buy a house, in that case people might increasingly neglect risk language to improve their chance of buying a house successfully. The last hypothesis is:
H5: The strength of the negative effect of risk language in real estate descriptions on selling times
15 These previous researches drive the idea that in the real estate market an effect could be found for different linguistic styles in the descriptions of a property on Funda.nl on the selling time of this object. The researches were in different industries, but all found effects from linguistic styles on various outcomes.
Methodology
Dataset
The dataset which will be used is a dataset provided by dr. F.B.I. Situmeang from the
University of Amsterdam. The dataset includes the data from 209339 properties that have been sold in 2018 and that have been listed on Funda.nl. The data includes the following variables for every property: publication date on funda.nl, the location of the property, the selling price, the full description used on the properties’ Funda page, the selling time (amount of days between publication date and date of signing the agreement) and the LIWC output of the descriptions. The variables that will be used to answer the main question are selling time, selling price and the LIWC output. The dataset has to be corrected for some properties that the agreement was signed for before being published on the Funda website, these have a negative selling time.
To be able to answer H5 we needed to add the average prices per municipality to the
dataset. We got the average housing prices per municipality in 2018 from CBS and added this data to the existing dataset. As written at the formulation of hypothesis 5, Krainer (2000) wrote that high average selling prices indicate a higher market heath as prices rise because there are more buyers than sellers on the market. This is why average housing prices are used as a proxy for market heath.
16 Also, there are some properties from which the selling price is missing, these properties will be deleted from the dataset. The dataset is large enough to be able to delete some of the properties without affecting the overall outcome of the analysis.
Linguistic inquiry and word count (LIWC) – analysing content and linguistic style of a
text
Linguistic inquiry and word count are used to analyse the textual data from Funda.nl as LIWC is a validated way to analyse texts by counting words and assigning them to psychologically
meaningful categories (Tausczick & Pennebaker, 2010). LIWC was developed in the mid-1990s. Tausczick and Pennebaker (2010) wrote that within a few years after developing the program in two broad categories were formed which were: content words and style words. These words have
different psychological and psychometric properties. The content words basically form the content of a written or spoken message, these words consist generally out of nouns, regular verbs, and many adjectives and adverbs. The other category, style words, identify how somebody writes and what someone conveys with what he or she is writing. The style words are often mentioned as function words (Ludwig et Al., 2013). LIWC will be used to analyze the data that this research uses. The data that will be analyzed are the descriptions of all the real estate objects which were listed on Funda.nl in 2018 and sold in 2018. The output that LIWC gives are the proportions of each word category such as pronoun, verb and adverbs. The proportions are formed by the program by taking the amount of words in each category and divide this by the total amount of words multiplied by 100. For instance, if a sentence has 20 words, and 5 words are verbs, then the proportion of verbs would be 25 in the output of LIWC (Tausczick & Pennebaker, 2010).
17
Model
Dependent variable
The dependent variable will be the same for all 5 variables. The outcome on which the effect will be measured is selling time of the property. Selling time in days is measured on a ratio scale. Selling times will be measured in days and the variable in the dataset is named: ‘dateDiff’. Datediff in the dataset is the result of the time between ‘publicatieDatum’ (Publicationdate) and
‘datum_ondertekening’ (signing_date). For instance, publication date is 05-01-2018 and signing date is 29-01-2018 then the result in Datediff is 24, which is in days. Because of a negative Datediff 3406 rows were deleted using Alteryx. These properties were sold before being posted on Funda.
Independent variables
In order to evaluate the hypotheses, the linguistic psychological dimensions need to be
operationalized. These dimensions are: positive affective language, over-positive affective language and risk language.
To evaluate hypothesis 1 positive affective language will be used as input variable in the linear regression analysis. Positive affective language is operationalized with LIWC by taking the ratio of positive affective words in the total number of words in the description. LIWC gives one variable called ‘posemo’, which is the ratio of positive affective words in the text. The outcome variable will be ‘datedif’, which is the amount of days before a property is sold.
To evaluate hypothesis 2 the variable positive affective language will again be used as the input variable. Positive affective language will be used as input variable in the curve estimation analysis. Positive affective language is operationalized with LIWC by taking the ratio of positive affective words in the total number of words in the description. LIWC gives one variable called ‘posemo’, which is the ratio of positive affective words in the text. The outcome variable will be ‘datedif’, which is the amount of days before a property is sold.
18 To evaluate hypotheses 3, 4 and 5, we are searching for risk in real estate descriptions. Risk language will be operationalized with LIWC by taking the ratio of ‘certain’ words from the total number of words in the descriptions. LIWC gives the variable called ‘certain’. ‘Certain’ words give the amount of certainty in a description. Less certainty means more risk, that is why the effect has to be interpreted in the opposite direction. For hypothesis 3 ‘certain’ will be the independent variable and ‘datediff’ will be used as outcome variable.
For Hypotheses 1, 2 and 3 a multivariate regression will be executed using SPSS. The model includes 9 variables provided by LIWC, which are chosen on the basis of existing literature.
To assess the interaction of selling price with risk language for H4 the data provides us with ‘Koopprijs’ as the moderating variable. This is the price for which the property was bought by the buyer. The selling price is measured on a ratio scale. The variable contained 539 rows in which the selling price was ‘NULL’, these rows were deleted using Alteryx. The input variable will be ‘certain’ and the outcome variable is ‘datedif’. The interaction effect will be analysed using the PROCESS macro in SPSS (Hayes, 2013).
To assess the interaction effect of average selling prices with risk on selling time the average selling prices per municipality will be used. The date per municipality was recovered from the CBS website and added to the data set using Excel. This data shows the average selling price in 2018 per municipality and will be analysed using the PROCESS macro in SPSS (Hayes, 2018). The input variable used is ‘certain’, the moderating variable used is ‘Averagesellingprice_gemeente’ and the output variable used is ‘datedif’.
Control variables
The control variables that are used closely follows the existing literature on linguistic effects in crowdfunding. From Lee at al. (2019) and Allison et al. (2015) the following LIWC output variables are used as control variables: Friend, Word count (Lee et al. (2019) uses pitch length), discrepancy, perceptual, social, money and achievement language.
19
Multicollinearity
The model has been tested with the independent variables and the control variables for multicollinearity. The VIF values were all close to 1 (see table 1) which indicates that there is no multicollinearity.
(Table 1) VIF statistics
VARIABLE VIF VALUE POSEMO 1.364 CERTAIN 1.043 FRIEND 1.068 WC 1.071 DISCREP 1.176 PERCEPT 1.303 SOCIAL 1.259 MONEY 1.059 ACHIEVE 1.078
20
Results
The correlation table shows that except for two correlations (see table 2), all correlations are significant at the 0.01 level, but only two correlations are above .3 (bold in table 2). These two correlations are perceptual language with positive emotional language and social language with discrepancy language. These correlations are both between 0.3 and 0.49 which means that there is a moderate degree of correlation. The rest of the correlations are all between -.3 and .3 which means that there is no correlation.
The mean statistics in table 2 show that the average selling time of the 205393 sold houses is 62.17 days, approximately 2 months. The average description length of each house is 385.05 words and the highest average ratios of language categories used in the descriptions are positive emotional language and social language.
(Table 2) Correlations table
To test hypothesis 1, positive affective language used in real estate object descriptions reduce selling times, a multivariate regression was performed on the data. The model gives an adjusted R-squared of 0.011 which indicates that the model as a whole has low predictive value. The coefficient of ‘posemo’ is -2.590 and is significant with P<0,01. The output of the PROCESS macro (Hayes, 2018) which are analysed for H4 and H5 also indicate a reduction of selling time when the ratio of positive emotional language increases (see table 4 and table 6). The Process output (Hayes, 2018) for H4 gives a coefficient of -2.903 with P<0.01. The Process output (Hayes, 2018) for H5 gives
VARIABLE M SD 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 1.DATEDIF 62.17 70.191 1 2.POSEMO 2.561 1.355 -.038* 1 3.CERTAIN .965 .703 .025* .084* 1 4.FRIEND .103 .212 -.053* .064* .022* 1 5.WC 385.05 218.418 -.006* -.066* .054* .031* 1 6.DISCREP .325 .418 .024* .138* .165* .079* .040* 1 7.PERCEPT 1.766 1.104 -.031* .435* .059* -.006* .144* .019* 1 8.SOCIAL 2.755 1.893 .001 .185* .125* .029* .138* .322* .135* 1 9.MONEY .626 1.450 -.032* -.144* .008* .038* -.039* .099* -.131* .136* 1 10.ACHIEVE 1.038 .696 .047* .199* .098* .086* .017* .182* .058* .092* -.006* 1
21 a coefficient of -2.522 with P<0.01. So H1 is supported as there is a significant effect of positive affective language on houses’ selling times.
(table 3)Regression analysis results
Dependent variable = dateDif
To test hypothesis 2, Over-positive affective (Graph 1) Curve estimation language used in real estate object descriptions
diminishes the positive effect of positive affective language on selling times, a curve estimation is performed on the data. The graph on the right (graph 1) shows an increasing quadratic line, but not a U shaped line, so H2 is rejected.
To test hypothesis 3, risk in real estate descriptions increase selling times, the same output of the earlier multivariate regression is used. The coefficient of ‘certain’ is 2.175 with P<0.01. As certain language is the opposite of risk language, the coefficient should be interpreted in the opposite direction, so 1% increase in the risk language ratio means an decrease of 2,175 days in selling time.
UNSTANDARDIZED B COEFFICIENTS STD. ERROR STANDARDIZED COEFFICIENTS BETA T SIG. (CONSTANT) 64.865 .523 123.991 .000 POSEMO -2.590 .133 -.050 -19.498 .000 CERTAIN 2.175 .224 .022 9.720 .000 FRIEND -19.877 .753 -.060 -26.407 .000 WC (WORD COUNT) -.004 .001 -.012 -5.153 .000 DISCREP 3.421 .406 .020 8.435 .000 PERCEPT -1.386 .159 -.022 -8.702 .000 SOCIAL .782 .091 .021 8.567 .000 MONEY -2.120 .109 -.044 -19.396 .000 ACHIEVE 5.639 .230 .056 24.533 .000
22 The output of the PROCESS macro (Hayes, 2018) which are analysed for H4 and H5 do indicate an increase of selling time when the ratio of risk language increases (see table 4 and table 6). The Process output (Hayes, 2018) for H4 gives a coefficient of -62.119 with P<0.01. The Process output (Hayes, 2018) for H5 gives a coefficient of -89.814 with P<0.01 which is also significant. These are the coefficients for certain language, which need to be interpreted in the opposite direction, so these coefficients do indicate that risk language increases selling times. On the basis of the PROCESS outputs H3 would be supported, but in the overall model used to test H3 the hypothesis is not supported. For now H3 is therefor not supported and needs further research for a clear conclusion.
(Table 4) PROCESS matrix output interaction effect certain x buyplog
Coeff
SE
T
P
LLCI
ULCI
Constant
-211.8187 6.8403 -30.9663 .0000 -225.2255 -198.4119Certain
-62.1189 5.6623 -10.9707 .0000 -73.2168 -51.0210Buyplog
52.4000 1.2667 41.3679 .0000 49.9174 54.8827INT_1
11.8350 1.0412 11.3663 .0000 9.7942 13.8758Posemo
-2.9031 .1307 -22.2126 .0000 -3.1593 -2.6470Friend
-19.5424 .7405 -26.3903 .0000 -20.9938 -18.0910WC
-.0226 .0008 -29.8967 .0000 -.0241 -.0211Discrep
3.8542 .3990 9.6599 .0000 3.0722 4.6362Percept
-2.3322 .1571 -14.8464 .0000 -2.6401 -2.0243Social
1.1653 .0900 12.9520 .0000 .9890 1.3416Money
-1.5230 .1078 -14.1332 .0000 -1.7342 -1.3118Achieve
4.6200 .2264 20.4047 .0000 4.1763 5.0638 Dependent variable: dateDifTo test hypothesis 4, The strength of the negative effect of risk language in real estate descriptions on selling times become less strong as housing prices rise, the PROCESS macro (model 1) of Hayes (2018) is used. ‘Certain’ is used as predictor variable and ‘Datedif’ is used as the outcome variable. The covariates used were the 8 LIWC output variables. As SPSS was bugged and could not run the PROCESS macro (Hayes, 2018) with the variable ‘KoopPri’ as the values were too big, the ‘KoopPri’ was computed into a new variable which is the logarithmic variable of ‘koopPri’ named ‘buyplog’. The ‘buyplog’ is used as the moderator. The outcomes should be interpreted in the opposite direction as certain language is the opposite of risk language. The outcome shows an
23 interaction effect which is significant with P<.001 (see table 4). So, there is an interaction effect in the model. Looking at the conditional effects of the focal predictor at values of the moderator (see
table 5) it shows that hypothesis 4 (Table 5) Conditional effects of ‘certain’ under buyplog is rejected. At a value of buyplog=5,2430
the effect of ‘certain’ language on selling
times is -.0674 with a P>0.05 (P=0.820), so this is no significant effect. At a value of buyplog= 5.439 the coefficient of ‘certain’ on ‘datedif’ is 2.246 with P<0.001 which is significant P=0.05). At a value of buyplog=5.652 the coefficient of ‘certain’ on ‘datedif’ is 4.777 with P<0.001 which is significant. Again, as certain language is the opposite of risk language the coefficients need to be interpreted in the opposite direction. H3 already rejected the negative effect of risk language on selling time, in table 3 again there is no negative effect to be seen, but what can be seen is that for higher selling prices, the effect of risk language on the selling time becomes significant and positive, so a shorter selling time. Therefor you see that with the increase of the house price, the negative effect of risk language diminishes and risk language even reduces the selling time at a certain price. So H4 is rejected, but there is room for further research on the effect that is found.
(Table 6) PROCESS matrix output interaction effect certain x avsellog
Dependent variable: dateDif
BUYPLOG EFFECT P
5.2430 -.0674 .8202
5.4385 2.2464 .0000
5.6522 4.7756 .0000
Coeff
SE
T
P
LLCI
ULCI
Constant
469.0731 13.5626 34.5859 .0000 442.4908 495.6554Certain
-89.8139 11.5906 -7.7488 .0000 -112.5313 -67.0966Avsellog
-74.6037 2.4946 -29.9060 .0000 -79.4930 -69.7143INT_1
16.7681 2.1310 7.8687 .0000 12.5914 20.9447Posemo
-2.5222 .1323 -19.0579 .0000 -2.7815 -2.2628Friend
-18.6001 .7507 -24.7764 .0000 -20.0715 -17.1287WC
.0019 .0007 2.5600 .0105 .0004 .0033Discrep
2.9846 .4042 7.3835 .0000 2.1923 3.7769Percept
-1.0013 .1589 -6.3002 .0000 -1.3128 -.6898Social
.9378 .0912 10.2848 .0000 .7591 1.1165Money
-2.1197 .1089 -19.4583 .0000 -2.3332 -1.9062Achieve
5.3430 .2291 23.3194 .0000 4.8940 5.792124 To Test H5, the strength of the negative effect of risk language in real estate descriptions on selling times become less strong as the heat of the market rises, the PROCESS macro (model 1) of Hayes (2018) is used. ‘Certain’ is used as predictor variable and ‘Datedif’ is used as the outcome variable. The covariates used were the 8 LIWC output variables.
As SPSS was bugged and could not run the PROCESS macro (Hayes, 2018) with the variable ‘Avselpr’ as the values were too big, the ‘Avselpr’ was computed into a new variable which is the logarithmic variable of ‘Avselpr’ named ‘Avsellog’. The ‘Avsellog’ is used as the moderator. The outcomes should be interpreted in the opposite direction as certain language is the opposite of risk language. The outcome shows a significant interaction effect with P<0.01 (see table 6). The
coefficient of the interaction effect is 16.768. Looking at the conditional effects of the focal predictor at values of the moderator (see table 7) it shows that hypothesis 5 is rejected. This is firstly because there is no significant negative (Table 7) Conditional effects of ‘certain’ under avsellog
effect of risk language on the selling time (see H3) so this
negative effect cannot be diminished either. Looking at the conditional effects of the focal predictor at values of the moderator you can see that at the lowest displayed value of avsellog, 5.356, there is no significant effect (P>0.05) with a P value of .964. But if you look at the next value of avsellog, 5.4363, then there is a significant effect to be found with P<0.001 and a coefficient of 1.343. This means that at a higher price a significant interaction effect has been evolved which causes the selling time to decrease. Looking at the highest displayed avsellog, 5.540, a significant interaction effect is also to be found with P<0.001 and a coefficient of 3.078. This indicates that the positive effect that was discovered at the previous value of the avsellog became stronger. So when market heat increases, a higher risk ratio in the descriptions cause a faster selling time. H5 is rejected, but this positive effect is again something for further research.
AVSELLOG EFFECT P
5.3555 -.0128 .9640
5.4363 1.3430 .0000
25
Discussion
This study was executed to analyse the effects of linguistic styles on selling times in the real-estate market.
The results that can be taken from this research are that certain linguistic styles used in house descriptions can affect the selling time of houses. In line with Lee et al. (2019) and Allison et al. (2017) the study shows that as somebody who tries to sell a house increases the ratio of positive affective language in the description the selling time decreases. The regression analysis shows that an increase of 1 percent in the ratio of positive emotional language in the description, results in a decrease of 2.590 days in selling time of the house. So, hypothesis 1 that proposes positive affective language used in real estate object descriptions reducing time to sell, is supported. This result could be caused by the phenomenon that was written about by Juanchich et al. (2012) in which he wrote that humans tend to perceive positive messages as a likelihood-communication device. This outcome is important as a bigger understanding for house sellers of what increases and what decreases the selling time of a property, might help then selling properties quicker. For real-estate agents using more positive emotional language could result in the same turnover from selling a house in a smaller period which could improve their overall income.
However, the analysis did not support hypothesis 2. This means that there was no proof found that over-positive emotional language would increase selling time as the description would be perceived less credible. This result is not in line with the existing literature on gathering investments from Angel investors from Parhankangas and Ehrlich (2014). A possible explanation might be that there has been no over-emoting in the data set and that all description writers are careful with not using too much positive emotional language in their descriptions.
Also, the study does not give unambiguous results for hypothesis 3. The regression analysis of hypothesis 3 does not support that risk language has a negative overall effect on the selling time
26 of a house, but the PROCESS outputs do indicate that this negative effect exists. Further research should look at what causes the difference between these effects.
The study also shows on the basis of hypotheses 4 and 5 that the use of risk language can decrease selling times under specific circumstances. These circumstances are both connected to the price of the houses. The research shows that as house prices increase the effect of risk language is that it decreases selling time at a certain point. Also, if market heath is higher the use of risk language also seems to decrease selling times. A possible explanation for this can be that more wealthy people are less risk-averse like Riley & Chow (1992) found in their study. These two findings need further research before a clear conclusion can be drawn on them. Further research could search for the price at which risk language actually decreases selling times.
This study does partially disagree to previous research. Lee et al. (2019) and Allison et al. (2015) found that in crowdfunding projects risk language reduced the funding success. Hypothesis 3 did not show a significant negative effect of risk language on the selling times, and hypotheses 4 and 5 even showed a positive effect of risk language on the time to sell at certain selling prices. But, as stated earlier: the results of H3, H4 and H5 do not indicate the same effect for risk language on selling times. So, the findings of Lee et al. (2019) and Allison et al. (2015) are not fully applicable in the real-estate market as they are in a crowdfunding environment.
There are some limitations to this research. For hypothesis 5 the market heath has been used as a moderator. Market heath has been defined as the average selling price per municipality, this is a good indicator, but other factors as amount of transactions also influence market heath. The specific market heath statistics were not available per region, that is why the average selling price had to be used, but market heath data per municipality could have given more accurate results. A second limitation is that 2018 has been analysed as one period, no distinction has been made between different quarters while for instance market heath might also have changed within different
27 quarters. As this study is not focused at these periodic differences in the real-estate market, it is not a problem for the study, but the differences could be included in the analyses of future research.
Further research could look at the results of linguistic styles on selling outcomes. So in stead of researching the time it takes to sell a house, it could analyse if for instance a higher ratio of positive affective language increases the relative selling price of a house. The current research shows that positive affective language reduces selling time, but it does not look at the results of the sale, which in many cases is also very important. Further research could also look at the analysis of H4 and H5 that show a positive effect of risk language on selling times under higher selling prices, why is this result to be seen and is this significant?
Conclusion
Previous research on the influence of linguistic style on different successes focused on crowdfunding projects and left unclear if these results were also extendable to the real-estate market. Thus, his research was conducted to gain insight on the effects of linguistic styles on selling times in the real-estate market. The research question that is posed for this research is: ‘Does the linguistic style of a housing description affect the selling time of a house which is for sale?’. The simple answer to this question is: yes, linguistic styles do affect the selling times of a house which is for sale. This study shows that the use of positive affective language has a positive effect on selling times of a house. Further on, the study shows that risk language might have a negative effect on selling times of a house, but this has to be further researched as different results were found on this effect. Lastly, the study found a positive effect of risk language on the selling time at higher selling prices, further research has to be done to validate this effect. In short, the outcome of this study need to be further researched before being widely useable for people like real-estate agents. The practical implication which the study does give, is that house sellers should try to use positive emotional language in their housing descriptions to decrease the time to sell their house.
28
Bibliography:
Allison, T. H., Davis, B. C., Short, J. C., & Webb, J. W. (2015). Crowdfunding in a prosocial microlending environment: Examining the role of intrinsic versus extrinsic cues. Entrepreneurship Theory and Practice, 39(1), 53-73.
Allison, T., Davis, B., Webb, J., & Short, J. (2017). Persuasion in crowdfunding: An elaboration
likelihood model of crowdfunding performance. Journal of Business Venturing, 32(6), 707–725. https://doi.org/10.1016/j.jbusvent.2017.09.002
Clayton, J., Mackinnon, G., & Peng, L. (2008). Time Variation of Liquidity in the Private Real Estate Market: An Empirical Investigation. The Journal of Real Estate Research, 30(2), 125–160.
Comparing LIWC2015 and LIWC2007. (n.d.). Retrieved from https://liwc.wpengine.com/compare-dictionaries/
Connelly, B., Certo, S., Ireland, R., & Reutzel, C. (2011). Review of Signaling Theory: A Review and Assessment. Journal of Management, 37(1), 39–67. https://doi.org/10.1177/0149206310388419 Dammingh, J. J. (2018). Makelaardij 2015–2017.
Hayes, A. F. (2017). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. Guilford publications.
Juanchich, M., Sirota, M., & Butler, C. (2012). The perceived functions of linguistic risk quantifiers and their effect on risk, negativity perception and decision making. Organizational Behavior and
Human Decision Processes, 118(1), 72–81. https://doi.org/10.1016/j.obhdp.2012.01.002
Krainer, J. (2001). A Theory of Liquidity in Residential Real Estate Markets. Journal of Urban Economics, 49(1), 32–53. https://doi.org/10.1006/juec.2000.2180
Kumar, M. D., Kambuaya, M., Jamil, F., & Muneer, S. (2015). Determinants of the risk tolerance of individual investors. International Journal of Economics and Financial Issues, 5(1S), 373-378.
29 Lee, C., Bian, Y., Karaouzene, R., & Suleiman, N. (2019). Examining the role of narratives in civic
crowdfunding: linguistic style and message substance. Industrial Management & Data Systems, 119(7), 1492–1514. https://doi.org/10.1108/IMDS-08-2018-0370
Ludwig, S., Ruyter, K., Friedman, M., Bruggen, E., Wetzels, M., & Pfann, G. (2013). More than words: the influence of affective content and linguistic style matches in online reviews on conversion rates.(Report). Journal of Marketing, 77(1), 87–103. https://doi.org/10.1509/jm.11.0560 MacMillan, I. C., Siegel, R., & Narasimha, P. S. (1985). Criteria used by venture capitalists to evaluate
new venture proposals. Journal of Business venturing, 1(1), 119-128.
Novy‐Marx, R. (2009). Hot and Cold Markets. Real Estate Economics, 37(1), 1–22. https://doi.org/10.1111/j.1540-6229.2009.00232.x
Over Funda. (n.d.). Retrieved from https://content.funda.nl/over/funda/
Parhankangas, A., & Ehrlich, M. (2014). How entrepreneurs seduce business angels: An impression management approach. Journal of Business Venturing, 29(4), 543–564.
https://doi.org/10.1016/j.jbusvent.2013.08.001
Riley, W., & Chow, K. (1992). Asset Allocation and Individual Risk Aversion. Financial Analysts Journal, 48(6), 32. Retrieved from http://search.proquest.com/docview/1347930973/
Sjostrom, T., Corsi, A. M., & Lockshin, L. (2016). What characterises luxury products? A study across three product categories. International Journal of Wine Business Research.
Spence, M. (1973). Job Market Signaling. The Quarterly Journal of Economics, 87(3), 355–374. https://doi.org/10.2307/1882010
Stephens, K., Waller, M., & Sohrab, S. (2019). Over-emoting and perceptions of sincerity: Effects of nuanced displays of emotions and chosen words on credibility perceptions during a crisis. Public Relations Review, 45(5). https://doi.org/10.1016/j.pubrev.2019.101841
30 Tausczik, Y. R., & Pennebaker, J. W. (2010). Psychological Meaning of Words: LIWC and Computerized
31
