A systematic review comparing two popular methods to assess a Type D personality effect

Introduction: Type D personality, operationalized as high scores on negative affectivity (NA) and social inhibition (SI), has been associated with various medical and psychosocial outcomes. The recent failure to replicate several earlier findings could result from the various methods used to assess the Type D effect. Despite recommendations to analyze the continuous NA and SI scores, a popular approach groups people as having Type D personality or not. This method does not adequately detect a Type D effect as it is also sensitive to main effects of NA or SI only, suggesting the literature contains false positive Type D effects. Here, we systematically assess the extent of this problem. Method: We conducted a systematic review including 44 published studies assessing a Type D effect with both a continuous and dichotomous operationalization. Results: The dichotomous method showed poor agreement with the continuous Type D effect. Of the 89 significant dichotomous method effects, 37 (41.6%) were Type D effects according to the continuous method. The remaining 52 (58.4%) are therefore likely not Type D effects based on the continuous method, as 42 (47.2%) were main effects of NA or SI only. Conclusion: Half of the published Type D effect according to the dichotomous method may be false positives, with only NA or SI driving the outcome.


Introduction
Type D ("Distressed") personality has been related to various medical and psychosocial outcomes, such as the occurrence of major cardiac events [1,2], depression, and anxiety [3].The construct Type D personality is hypothesized to affect these outcomes through the combined influence of its two subcomponents, the personality traits negative affectivity (NA) and social inhibition (SI).Negative affectivity refers to the tendency of experiencing negative thoughts, feelings and emotions, while socially inhibited people experience difficulty in expressing these emotions and feelings in social situations [4].
Initially, Type D research mainly focused on how the combined influence of high NA and SI scores affects the prognosis of cardiovascular disease patients [5].These studies mainly involved hard endpoints, such as mortality and various cardiac events.Although some earlier studies on cardiovascular disease patients have not been replicated in subsequent research [6], one meta-analysis found support for a Type D effect on adverse events in CAD patients, but not in heart failure patients [7].Another explanation for the inconsistent findings involves differences between studies in the sample characteristics and studied endpoints [8].The Type D effect is arguably less pronounced in older patients and mortality endpoints, because at older ages, various medical comorbidities may become more important in explaining mortality than personality effects like Type D. Although such characteristics may in part explain why the Type D effects in studies involving older participants or heart failure patients could not be replicated, here we show that the methodological issue of how to operationalize Type D personality may also play an important role in explaining these inconsistencies.
A considerable debate exists on how the NA and SI traits combine in exercising a Type D effect [9][10][11][12].An additive effect would mean that the Type D effect is equal to the sum of the separate NA and SI effects, while a synergistic effect would imply that the Type D effect is more than the sum of its parts.Both additive and synergistic Type D effects can best be modeled by including both the continuous NA and SI scores and their interaction as predictors in a regression analysis [13].Although several authors have advocated the use of this continuous method [9][10][11][12][13], it has remained more common to assess the Type D effect using a dichotomous operationalization.According to this dichotomous method, people are classified as having a Type D personality when they score above a predetermined cut-off on both NA and SI.In some studies, the dichotomous method is extended to a categorical method with four personality groups, which further divides the people without Type D personality in three groups based on the different combinations of scoring above or below the cut-off on the two traits: Historically, the development of the dichotomous method was motivated by clinical and empirical considerations.Using cluster analysis, Denollet and colleagues [5] detected a clinical discontinuity between people with or without high scores on both NA and SI.The prevalence of this Type D personality type was shown to be higher in people suffering from cardiovascular disease than in healthy controls [5].In these earlier studies, NA and SI were assessed with the trait anxiety scale and a social inhibition subscale of the heart patient's psychological questionnaire.Inspired by these measures, Denollet developed the Type-D scale-16 (DS16) [14], measuring each of the NA and SI constructs with eight items.Seven years later the DS16 was revised into the slightly shorter DS14 instrument that provided a more balanced assessment of the various aspects of the NA and SI constructs [4].From 2005 onwards, the DS14 became the standard instrument used to assess Type D personality.
In the early Type D studies, researchers exclusively used the dichotomous and categorical methods that classified people in two or four personality groups, either based on a predetermined cut-off score of 10 (in case of the DS14) or a median split (in case of other measurement instruments).These methods have been criticized in several studies based on conceptual and empirical arguments [9][10][11][12].Although Whitehead and colleagues [15] were the first to use a continuous method to estimate a Type D effect, Ferguson and colleagues [9] were the first to explicitly argue that Type D personality can better be conceptualized and analyzed as a continuous construct.Several years laterff, Smith [10] warned that the dichotomous and categorical methods could produce spurious Type D effect.Recently this was confirmed based on various computer simulations investigating the adequacy of these methods in estimating the Type D effect [11,13].
These simulations showed that the dichotomous and categorical methods often fail to detect the Type D effect adequately, because they often tend to produce false-positive Type D effect (i.e.Type I errors) when only NA or SI is related to the outcome.Although these personality group methods are sensitive to any kind of NA or SI effect, at the same time they are less powerful in detecting a particular significant effect than the continuous method and may therefore also produce more falsenegative findings (i.e.Type II errors).Reducing the continuous NA and SI measures to two or four personality types reduces the information about individual differences on these personality traits and this practice is associated with a loss in statistical power up to 60% [16].The simulation studies also indicated that a correctly specified continuous model does not suffer from this problem and is able to correctly identify the underlying NA and SI effects [11,13].
The bias of the dichotomous and categorical approaches is not limited to research on Type D personality, but to any field where two continuous measures are transformed into a variable indicating whether or not someone scores above a cut-off on both measures.Examples are defensive hostility (high scores on both defensiveness and hostility [17]), mixed states in bipolar disorder (high scores on both mania and depression [18]), or androgynous gender schemas (high masculinity and femininity gender scores [19]).When the two measures are correlated, the dichotomous and categorical approaches overestimate the presence of Type D effects.When the measures are uncorrelated, only the dichotomous method shows such positive bias, yet the categorical method using 4 groups still results in lower power [11,13].
Given that most of the studies in the Type D literature have used the dichotomous method, the conclusions drawn from significant dichotomous effects may have to be reconsidered.In these studies, Type D personality may not be responsible for explaining individual differences in the dependent measure, but rather NA or SI only.However, the extent of this bias remains unclear.It would therefore be interesting to know the percentage of studies in which the dichotomous method and continuous method lead to different conclusions.To determine the extent to which the dichotomous and continuous methods have produced different conclusions in the Type D literature, one would ideally like to compare the results of both methods in all published empirical studies on Type D personality.Unfortunately, most published studies only used one method to assess the Type D effect, partly because the continuous method has only been argued for in the literature from 2009 onwards [9].However, it is still possible to investigate the differences in findings between the dichotomous and continuous method, in the subset of studies that reported the results according to both methods.This is the aim of the present study.
Here, we present results of a systematic review of the empirical Type D literature, including any kind of Type D study as long the results were reported according to both the dichotomous and continuous methods.This is no traditional systematic review, as its purpose is not to answer the substantive question of whether Type D personality is related to a particular outcome.Its purpose is rather to assess how often the conclusions drawn from the continuous method and the dichotomous method differ.Based on these comparisons we can estimate the percentage of significant dichotomous effects that do not represent Type D effects (additive or synergistic) according to the continuous method.In line with earlier simulation studies [11,13], we expect that the dichotomous method produce more significant effects than the continuous method.As these simulations also indicated that the dichotomous method is sensitive to main effects of NA or SI only, we expect that some of the significant dichotomous effects will not be Type D effects according to the continuous method, but main effects of NA or SI.

Inclusion and exclusion criteria
We included studies that reported the effect of Type D personality on any dependent measure according to both the dichotomous method and the continuous method.Only studies written in the English language were included.We excluded studies using the continuous method if they analyzed the effects of NA and SI in separate univariable analyses.When only one of these traits is related to an outcome, univariable analyses risk producing significant effects for both traits, because the effect of one trait spills over to the other trait due to the moderate correlation between NA and SI.Studies were also excluded if their continuous model did not include the NA and Si main effects and their interaction.

Search strategy
The literature search was performed on November 4th 2019.The electronic databases Pubmed and PsycINFO were searched for articles containing the term 'Type D personality' in either the title, keywords, or abstract.The hits provided by these databases were filtered to select only empirical studies, resulting in 569 unique studies.Two independent researchers screened the full texts of these studies to determine whether they met the inclusion criteria listed above.This turned out to be the case for 36 studies.Most excluded studies did either not use two methods to assess the Type D effect, or used statistical analyses that did not meet our inclusion criteria (e.g.zero-order correlations).A cited reference search of Type D's measurement instrument, the DS14 questionnaire [4] was performed using Web of Science and resulted in 665 P. Lodder et al. studies, of which 398 were not yet identified by the earlier search.Of these 398 studies most did not investigate a Type D effect, yet eight not yet identified studies met our inclusion criteria, resulting in a total of 44 included studies.

Data extraction
For each study, two researchers independently extracted the following data: [1] First author's name; [2] Journal; [3] Publication year; [4] Sample size; [5] All dependent measures where Type D personality was used as a predictor; [6] For each dependent measure, the reported effect size (e.g.OR, HR, R 2 , Beta) or test statistic (e.g.t, F, Z) and p-value according to the dichotomous method and the [7] continuous method.The effect size was preferred if both an effect size and test statistic was reported.If neither an effect size nor the test statistic was reported, the effect was considered missing and only the p-value was extracted.When articles reported the effect size with a 95% confidence interval instead of a p-value, we calculated the p-value based on the standard error extracted from the confidence interval.If p-values and confidence intervals were not reported, statistical significance was either determined based on whether the authors reported the effect as statistically significant, or else was considered missing.Main effects in models without interactions were preferred over main effects in models including the interaction.Adjusted effects were preferred over unadjusted effects.

Data analysis
For both the dichotomous method and continuous method, the frequency and percentage of significant p-values was calculated.P-value distributions were visualized using histograms.Cohen's kappa was used to determine the agreement between the conclusions drawn from both methods.
We did not conduct standard meta-analytic tests on the extracted data because the purpose of this project was not to aggregate the estimated effect sizes of studies in the Type D literature.Such aggregates would not be very meaningful as the included studies almost never focused on similar outcome measures.Our aim was merely to compare the conclusions drawn from two methods commonly used to assess the Type D effect, given the typical sample sizes and statistical power encountered in the Type D literature.For this reason, the known limitations of vote counting [20] are not relevant to the present study.Vote counting involves counting the statistically significant p-values and this practice can be problematic because studies that are underpowered may produce non-significant effects, even when there are real effects underlying the data [21].
The power to detect significant effects differs between the dichotomous and continuous methods.Computer simulations show that continuous methods in general have more statistical power than methods using dichotomized variables [11,13,22].In Appendix A we report the results of a small simulation study indicating that the continuous method always showed more statistical power to detect a Type D effect than the dichotomous method.This shows that when the dichotomous method results in a significant effect and the continuous method in a non-significant effect, then this difference is likely not explained by a lower statistical power of the continuous method.An alternative explanation could be that the dichotomous method is sensitive to any kind of NA or SI effect (main/quadratic/interaction), whereas the continuous method adequately detects the presence of each of these different types of effects.A1 presents 14 published studies that were excluded from our review because the continuous method was not modeled appropriately.The authors of these studies took seriously the recommendation to analyze Type D personality continuously by investigating the interaction between NA and SI (except for one study where the sum of NA and SI rather than their product was investigated).However, in these excluded studies, these products (or sums) were used in subsequent analyses without adjusting for the continuous NA and SI main effects.As a result, these analyses may suffer from a problem similar to that of the dichotomous approach: they cannot distinguish between four kinds of underlying effects: [1] NA main effect; [2] SI main effect; [3] Additive Type D effect, or [4] Synergistic Type D effect.Because these studies did not meet our inclusion criteria due to not using the correct continuous method, we have excluded them from further analyses.

Main findings
The flowchart in Fig. 1 indicates that of all 967 empirical studies, 44 (7.7%) were included in our review.All included studies were published after 2009, the year when the first recommendation to assess the Type D effect with a continuous method was published.Together, the 44 included studies investigated 158 effects of Type D personality on a dependent measure.Each of those 158 effects was assessed using both the dichotomous and continuous method.
Table A2 in Appendix C shows for each study included in our review, the estimated Type D effect and p-value according to the dichotomous and continuous method.Table 1 summarizes these findings by presenting for both the significant and non-significant dichotomous effects the percentage of significant effects according to the continuous method.Of all 158 effects, 89 (56.3%) were significant according to the dichotomous method, 67 (43.8%) effects concerned a significant continuous NA effect, 35 (22.9%) a significant continuous SI effect, 16 (10.5%)an additive Type D effect (NA + SI), and 26 (16.5%) a synergistic Type D effect (NA*SI).The direction of all but one [36] of these synergistic Type D effects was in line with what Type D theory would predict.
To determine the agreement in statistical significance between the dichotomous and the continuous methods in assessing the Type D effect, both Cohen's kappa and the percentage of agreement were calculated.It turned out that the Type D effect assessed according to the dichotomous Fig. 1.Flowchart of studies included in the systematic review.
P. Lodder et al. method showed poor agreement with both the continuously assessed SI effect (κ = 0.19; 95%CI = 0.08, 0.31; agreement = 56.2%),yet reasonable agreement with the NA effect (κ = 0.55; 95%CI = 0.42, 0.68; agreement = 77.1%).The dichotomous method showed poor agreement with the additive Type D effect (κ = 0.14; 95%CI = 0.06, 0.21; agreement = 51.6%), and even worse agreement with the synergistic Type D effect (κ = 0.08; 95%CI = − 0.03, 0.18; agreement = 50.0%).This makes sense, as earlier research has shown that the dichotomous method is not so much sensitive to additive or synergistic Type D effects, but more to the presence of any NA or SI effects.Indeed, the dichotomous method showed the best agreement with the detection of any continuous effect (i.e.NA or SI main effect or their interaction: κ = 0.59; 95%CI = 0.46, 0.72; agreement = 80.4%).These results indicate that the dichotomous method is very sensitive, but not very specific in detecting the kind of underlying effects.Supplemental Table 1 shows the results of sensitivity analyses for different types of outcomes.Cohen's kappa could not be estimated for the mortality outcomes due to low cell counts in the cross table.Nevertheless, regardless of whether researchers studied the cardiometabolic or psychosocial outcomes, the results were similar to those of the overall analysis.
For all studies included in the systematic review, Fig. 2 shows the pvalue distributions according to both the dichotomous method and the main-and interaction effects resulting from the continuous method.The presence of true effects is indicated by right skewed p-value distributions, with a higher chance on observing lower p-values (e.g., 0.01) than high (e.g., 0.04) p-values.Under the truth of the null hypothesis the pvalue distribution is expected to be uniform, with an equal chance on observing any p-value [77,78].On first sight, the distributions of the dichotomous method and the continuous NA effect look rather similar.These distributions are both very right skewed and therefore indicate a larger evidential value than the distributions of the continuous SI effect

Table 1
For all studies included in our review, the number (%) of statistically significant results according to the continuous (rows) and dichotomous (columns) methods.Fig. 2. For all effects included in the systematic review, the frequency distribution of observed p-values resulting from the dichotomous method and the three effects of the continuous method.The dotted line indicates a significance level of 0.05.
P. Lodder et al. and the interaction between NA and SI.Interestingly, the p-value distribution of the interaction effect looks most uniform of all, suggesting the least evidential value for synergistic Type D effects.Given that the continuous method is much more specific than the dichotomous method in identifying the type of underlying effects, it would be interesting to evaluate the results of the continuous method within the subset of significant dichotomous effects.This would help explain in what way the NA and SI personality traits influence a dependent measure whenever the dichotomous effect is significant.Fig. 3 shows for all significant dichotomous effects the percentage of significant effects according to the continuous method.It turned out that of the 89 significant dichotomous effects, 15 (16.9%) were found to be additive Type D effects, and 22 (24.7%) were synergistic Type D effects.Assuming that a Type D effect is either additive or synergistic, 41.6% of the significant dichotomous effects were Type D effects according to the continuous method.The remaining 58.4% significant dichotomous effects method (55 effects) are therefore likely not Type D effects based on the continuous method.
The most frequently observed result in the continuous analyses was a significant effect for NA only in 34 (38.2%) of the significant dichotomous effects.In 8 (9.0%) of the significant dichotomous effects only SI was related to the dependent measure.These results suggest that 47.2% of the significant dichotomous effects are in fact not caused by the combined influence of NA and SI (be it additively or synergistically), but rather by one of these two personality traits only.Lastly, for 10 (11.2%) of the significant dichotomous effects, no significant continuous effect was found.

Discussion
The purpose of this study was to determine the discrepancy in the results based on the continuous method and dichotomous method in assessing the Type D effect.Our analyses indicated that the dichotomous method shows poor agreement with both the continuously assessed additive and synergistic Type D effects.For the studies included in our review, the dichotomous method showed reasonable agreement with the NA main effect, suggesting that in many Type D studies only NA is sufficient in explaining variance in the dependent measure.
Earlier research [11,13] indicated that the dichotomous method is not very specific, because it is sensitive to the presence of any underlying NA or SI effect, including main effects, quadratic effects and interactions.This suggests that the results of published studies using only the dichotomous method should be reconsidered because these studies may have concluded that Type D personality is related to a dependent measure, while in reality the significant dichotomous effect could be caused by NA or SI only.For all dependent measures that are affected by only one of these two personality traits, the Type D personality construct is not necessary in explaining how people vary on the dependent measure [79].
The present study showed that 56.3% of the included dichotomous analyses showed statistically significant Type D effects.Of those significant effects, 58.4% were not Type D effects according to the continuous method.Assuming that the studies included in this review are representative (e.g., in terms of sample size and the kind of dependent measures studied) of all studies investigating a Type D effect, it can be concluded that almost 60% of the significant Type D effect reported according to the dichotomous method may be spurious Type D effects caused by the bias of the dichotomous method.Our review suggests that such spurious Type D effects are most likely explained by effects of NA only.These estimates should of course be interpreted with care as their generalizability is conditional on the representativeness of the studies included in our review.Our review included studies conducted from 2009 onwards, since at that point Ferguson and colleagues first argued to analyze Type D effects using the continuous method.Differences between included and excluded studies in terms of for instance study population (e.g.cardiac vs. healthy) or dependent measure (e.g.cardiac endpoints vs. mental health questionnaires) may have confounded our estimates.Nevertheless, our results suggest that at least part of the significant dichotomous effects reported in the Type D literature are likely main effects of NA or SI only.This highlights the importance that future research at least takes a closer look at this problem, not only in the context of Type D research, but also in other fields where two correlated continuous measures are dichotomized and transformed into subgroup variables.
Regarding future research on Type D personality, a first start would be to re-analyze the earlier published literature using the continuous method.Such analyses should not only investigate the NA and SI main effects, but also their interaction and their quadratic effects (see for instance Lodder et al. [2,3]).When testing interaction effects, it is important to check whether they are confounded by quadratic effects of the variables involved in the interaction, because not modeling quadratic effects when they are actually present may result in false positive interaction effects [13,80,81].Ideally, such re-analyses could be done separately for each type of outcome measure in the form of an individual patient data meta-analysis [82].Such meta-analyses combine the raw datasets of earlier published studies focusing on a similar research question.In the context of Type D personality this will allow for a sufficiently powered statistical test using the continuous method, to determine whether the earlier reported dichotomous effects are best explained by NA only, SI only, or the combined Type D effect (additive or synergistic).A first attempt to conduct such an analysis has already been initiated, investigating the Type D effect on adverse (cardiac) events in patients with coronary heart disease [83].
This study was motivated by earlier findings that the Type D effect can better be analyzed using a continuous approach.These studies [11,13] assumed a dimensional conceptualization of the Type D personality construct.A counterargument could be that the true mechanism underlying the Type D effect can better be seen as categorical, with a set of distinct latent personality classes giving rise to the different score patterns on the DS14 questionnaire [11].However, there appears to be a consensus that personality traits in general are dimensional in nature, raising the question why NA and SI would be an exception [9,84].Furthermore, based on a taxometric analysis, Ferguson and colleagues [9] showed that Type D can better be seen as a dimensional construct than as a categorical construct.Moreover, it is only a small step from pragmatically creating a categorical personality type variable to assigning concrete existence such artificially created categories, a process called reification.Nevertheless, we still believe in the utility of the label Type D personality, but more as a convenient description of a particular NA and SI score pattern, than as having an ontological reality of its own.
Our review also identified a set of 14 studies that did not appropriately use the continuous method to assess the Type D effect.In these Fig. 3.For all 89 significant Type D effects based to the dichotomous method, the percentage of observed Type D effects according to the continuous method.The black bars represent a 95% confidence interval.
studies, the Type D construct was operationalized as the sum or product of NA and SI scores.However, these sums/products were included in subsequent analyses without adjusting for the NA and SI main effects.Therefore, any significant Type D effects may be confounded by the presence of NA and/or SI main effects, making it unclear whether Type D personality is necessary in explaining individual differences in the dependent measure.These effects should be reconsidered in future research using the correctly specified continuous method.
For 11.2% of the significant dichotomous effects, no significant continuous effects were found.This finding is likely explained in terms of differences between the approaches in statistical power.Assuming that both NA and SI have a very small negative effect on a dependent measure and that the power to detect such effects is too low, resulting in non-significant continuous effects.Although dichotomizing continuous variables results in lower statistical power [16], the dichotomous effect may not necessarily have lower power than the continuous method, because it combines information from two dichotomized variables.Since the dichotomous method is sensitive to any underlying NA or SI effect, it may pick up explained variance from both the NA and SI main effects, producing effects large enough to be detected with sufficient power, even when the continuous tests of single personality traits are underpowered.
A strength of this study is that is the first to review the results of all published studies that have used both the continuous and dichotomous method to estimate a Type D effect.A comparison of the findings of those methods was necessary, as earlier simulations indicated that the dichotomous method could results in false positive Type D effects.Our study is the first to show that a major part of the significant dichotomous method effects published Type D literature is not a Type D effect according to the continuous method.However, a limitation of this review is that we could only include a small percentage of the Type D literature, because most studies did not use several methods to assess the Type D effect.A second limitation is that we only included published studies.Therefore, generalization of the current results to the entire Type D literature is conditional on the similarity between the included and excluded studies.
A third limitation is that in some included studies the Type D effect was assessed on more than one dependent measure.The Type D effects within these studies may be correlated, either because of they are based on the same sample of participants or because the investigated dependent measures are similar (e.g., the subscale scores of a multidimensional questionnaire).This could mean that the data used in our Cohen's kappa analyses were not independent, possibly resulting in biased estimates of the agreement between the dichotomous and continuous methods.The estimated percentage of false positive Type D effects in the literature is likely not affected by this limitation, assuming that this violation of independence is similar in the set of excluded studies.
A fourth limitation of this review is that the continuous methods did not test for the presence of quadratic NA or SI effects.Research shows that not including such quadratic effects in the statistical model when they are present, may cause spurious interaction effects when the two traits involved in the interaction are correlated, which is the case for NA and SI [13,81], but also for many other kinds of studies, such as those testing interaction effects between anxiety and depression [85].For Type D research, this implies that every reported significant interaction between NA and SI could be a quadratic effect of NA or SI only.It was not possible to investigate this issue in the current review, because, to our knowledge, only two published studies have investigated whether the synergistic Type D effect is confounded by the quadratic NA and SI effects [2,3].Future research should investigate the extent to which in the published literature unmodeled quadratic effects may have resulted in spurious synergistic Type D effects.
To conclude, this study showed that the majority of the significant dichotomous effects in the Type D literature are likely not additive or synergistic Type D effects, but rather main effects of NA or SI only.This stresses the importance of reconsidering all earlier studies on Type D personality that have used only the dichotomous method to assess the Type D effect.Although this paper focused on Type D personality, we hope our findings also motivate reanalysis of earlier results in other fields involving the combined effects of two correlated variables.These fields may contain many spurious findings if analyses were conducted using subgroup variables based on the dichotomization of two continuous variables.Reconsidering such earlier studies may shed light on why, in general, so many published findings are difficult to replicate.The use of adequate statistical methods minimizes the chance on incorrect conclusions (i.e., Type I & II errors) and is one way to increase the replicability of psychological science [86].

Fig. A1.
The percentage of significant Type D effects according to dichotomous method (red solid line) and continuous method (blue dashed line), varying over sample size, item skewness, the underlying NA main effect and NA*SI interaction effect.In all conditions, the size of the SI main effect was fixed to zero.

Appendix B. Important excluded studies
Table A1 shows a selection of 14 published studies that were excluded from our review.The authors of these studies took seriously the recommendation to analyze Type D personality continuously by investigating the interaction between NA and SI (except for one study where the sum of NA and SI rather than their product was investigated).However, in these excluded studies, these products (or sums) were used in subsequent analyses without adjusting for the continuous NA and SI main effects.As a result, these analyses may suffer from a problem similar to that of the dichotomous approach: they cannot distinguish between four kinds of underlying effects: [1] NA main effect; [2] SI main effect; [3] Additive Type D effect, or [4] Synergistic Type D effect.This problem may be circumvented by mean-centering the NA and SI variables before multiplying them [13].However, the studies in Supplemental Table 1 did not report that the NA and SI scores were mean-centered before their multiplication.One could argue that the scores were meancentered and the authors forgot to report it, but there are in fact several reasons to assume that raw scores rather than centered scores were used.First, several studies report the correlations between NA, SI and their product score [26,31,33].If NA and SI would have been mean-centered before multiplying them, these raw scores would barely correlate with their multiplication.However, the substantial correlations in these studies suggest that P. Lodder et al. raw scores were multiplied instead of the mean-centered scores.Second, several studies [30,32,35] show that the scale of the product variable was larger than expected.If NA and SI would have been mean-centered before multiplication, then the scale of their product would include both negative as well as positive values, with a mean close to zero.Based on the reported mean scores of two studies [32,35] and the plotted product scores of one study [30], it appears that the product scores only contain positive values with a mean high above zero.Together these results suggest that raw scores rather than mean-centered scores have been used when multiplying NA and SI.Using only the multiplied raw score in subsequent analyses, without adjusting the effect for the NA and SI main effects, results in biased estimates of the Type D effect (see Lodder [13] for empirical support for this bias).
As the studies in Supplemental Table 1 either did not investigate the NA and SI main effects, it remains unclear in what way the two Type D personality traits are related to the dependent measures.However, in some of these studies the bivariate correlation between both traits and the dependent measure were reported.For instance, Williams and colleagues [33] showed that the product of NA and SI was positively correlated with alcohol dependence.However, the bivariate correlations also indicate a significant positive correlation for SI with alcohol dependence, while the correlation with NA is very weak and not statistically significant.This pattern excludes the possibility of an additive Type D effect, or an NA only effect.Yet it remains unclear whether the underlying effect is caused by SI only or by an interaction between NA and SI above and beyond the SI effect.Because the studies listed in Supplemental Table 1 did not meet our inclusion criteria due to not using the correct continuous method, we have excluded them from our systematic review.

Table A1
Studies in which Type D personality (operationalized as NA+SI or NA*SI) was associated with an outcome while the analyses were not adjusted for NA and SI main effects.

Appendix C Table A2
For all studies included in the systematic review, the estimated Type D effect and p-value based on the dichotomous method and continuous method.P. Lodder et al.