Propensity score matching and variations on the balancing test

Balancing tests are diagnostics designed for use with propensity score methods, a widely used non-experimental approach in the evaluation literature. Such tests provide useful information on whether plausible counterfactuals have been created. Currently, multiple balancing tests exist in the literature but it is unclear which is the most useful. This article highlights the poor size properties of commonly employed balancing tests and attempts to shed some light on the link between the results of balancing tests and bias of the evaluation estimator. The simulation results suggest that in scenarios where the conditional independence assumption holds, a permutation version of the balancing test described in Dehejia and Wahba (Rev Econ Stat 84:151–161, 2002) can be useful in applied study. The proposed test has good size properties. In addition, the test appears to have good power for detecting a misspecification in the link function and some power for detecting an omission of relevant non-linear terms involving variables that are included at a lower order.

Identifying rate-limiting nodes in large-scale cortical networks for visuospatial processing: An illustration using fMRI

With the advent of functional neuroimaging techniques, in particular functional magnetic resonance imaging (fMRI), we have gained greater insight into the neural correlates of visuospatial function. However, it may not always be easy to identify the cerebral regions most specifically associated with performance on a given task. One approach is to examine the quantitative relationships between regional activation and behavioral performance measures. In the present study, we investigated the functional neuroanatomy of two different visuospatial processing tasks, judgement of line orientation and mental rotation. Twenty-four normal participants were scanned with fMRI using blocked periodic designs for experimental task presentation. Accuracy and reaction time (RT) to each trial of both activation and baseline conditions in each experiment was recorded. Both experiments activated dorsal and ventral visual cortical areas as well as dorsolateral prefrontal cortex. More regionally specific associations with task performance were identified by estimating the association between (sinusoidal) power of functional response and mean RT to the activation condition; a permutation test based on spatial statistics was used for inference. There was significant behavioral-physiological association in right ventral extrastriate cortex for the line orientation task and in bilateral (predominantly right) superior parietal lobule for the mental rotation task. Comparable associations were not found between power of response and RT to the baseline conditions of the tasks. These data suggest that one region in a neurocognitive network may be most strongly associated with behavioral performance and this may be regarded as the computationally least efficient or rate-limiting node of the network.

Association analysis of proopiomelanocortin (POMC) haplotypes in type 1 diabetes in a UK population

Aim

To assess the association of POMC haplotype-tagged single nucleotide polymorphisms (htSNPs) with the development of type 1 diabetes (T1D) in a Caucasian population.

Methods

All exons, intron 1, and approximately 6-kb upstream and 3-kb downstream of the POMC gene were bidirectionally resequenced to identify DNA polymorphisms in 30 individuals. Allele frequencies were determined (60 chromosomes) and efficient htSNPs were selected using the htSNP2 programme. Genotyping was performed in 390 cases, 339 controls and 245 T1D parent-offspring trios, using Taqman, Sequenom and direct-sequencing technologies.

Results

Thirteen polymorphisms (two novel) with a minor allele frequency greater than 1% were identified. Six POMC htSNPs (rs3754863 G>A, ss161151662 A>G, rs3754860 C>T, rs1009388 G>C, rs3769671 A>C, rs1042571 G>A) were identified. Allele and haplotype frequencies were similar between case and control groups (P>0.60 by permutation test), and assessment of allele transmission distortion from informative parents to affected offspring also failed to find any association. Stratification of these analyses for age-at-onset and HLA-DR risk group (DR3/DR4) revealed no significant associations. A haplotype block of 9.86-kb from rs3754863 to rs1042571 was identified, encompassing the POMC gene. Comparison of haplotype frequencies identified the GGCGAG haplotype as protective against T1D in 12.9% of cases vs. 18.3% of controls: ?2=8.18, Pc=0.03 by permutation test.

Conclusion

The POMC SNP haplotype GGCGAG may have a protective effect against T1D in the UK population. However, this finding needs to be replicated, and the cellular and molecular processes influenced by this POMC haplotype determined to fully appreciate its impact.

Genome scan meta-analysis of schizophrenia and bipolar disorder, part II:Schizophrenia

Schizophrenia is a common disorder with high heritability and a 10-fold increase in risk to siblings of probands. Replication has been inconsistent for reports of significant genetic linkage. To assess evidence for linkage across studies, rank-based genome scan meta-analysis (GSMA) was applied to data from 20 schizophrenia genome scans. Each marker for each scan was assigned to 1 of 120 30-cM bins, with the bins ranked by linkage scores (1 = most significant) and the ranks averaged across studies (R(avg)) and then weighted for sample size (N(sqrt)[affected casess]). A permutation test was used to compute the probability of observing, by chance, each bin's average rank (P(AvgRnk)) or of observing it for a bin with the same place (first, second, etc.) in the order of average ranks in each permutation (P(ord)). The GSMA produced significant genomewide evidence for linkage on chromosome 2q (PAvgRnk

Suggestive association with ocular phoria at chromosome 6p22

PURPOSE. We conducted a genome-wide association study to identify genetic factors that contribute to the etiology of heterophoria.

METHODS. We measured near and far vertical and horizontal phorias in 988 healthy adults aged 16 to 40 using the Keystone telebinocular with plates 5218 and 5219. We regressed degree of phoria against genotype at 642758 genetic loci. To control for false positives, we applied the conservative genome-wide permutation test to our data.

RESULTS. A locus at 6p22.2 was found to be associated with the degree of near horizontal phoria (P = 2.3 × 10 ). The P value resulting from a genome-wide permutation test was 0.014.

CONCLUSIONS. The strongest association signal arose from an intronic region of the gene ALDH5A1, which encodes the mitochondrial enzyme succinic semialdehyde dehydrogenase (SSADH), an enzyme involved in γ-aminobutyric acid metabolism. Succinic semialdehyde dehydrogenase deficiency, resulting from mutations of ALDH5A1, causes a variety of neural and behavioral abnormalities, including strabismus. Variation in ALDH5A1 is likely to contribute to degree of horizontal phoria.

A population study of binocular function

As part of a genome-wide association study (GWAS) of perceptual traits in healthy adults, we measured stereo acuity, the duration of alternative percepts in binocular rivalry and the extent of dichoptic masking in 1060 participants. We present the distributions of the measures, the correlations between measures, and their relationships to other psychophysical traits. We report sex differences, and correlations with age, interpupillary distance, eye dominance, phorias, visual acuity and personality. The GWAS, using data from 988 participants, yielded one genetic association that passed a permutation test for significance: The variant rs1022907 in the gene VTI1A was associated with self-reported ability to see autostereograms. We list a number of other suggestive genetic associations (p<10^-5).

Exact Nonparametric Two-Sample Homogeneity Tests for Possibly Discrete Distributions.

In this paper, we study several tests for the equality of two unknown distributions. Two are based on empirical distribution functions, three others on nonparametric probability density estimates, and the last ones on differences between sample moments. We suggest controlling the size of such tests (under nonparametric assumptions) by using permutational versions of the tests jointly with the method of Monte Carlo tests properly adjusted to deal with discrete distributions. We also propose a combined test procedure, whose level is again perfectly controlled through the Monte Carlo test technique and has better power properties than the individual tests that are combined. Finally, in a simulation experiment, we show that the technique suggested provides perfect control of test size and that the new tests proposed can yield sizeable power improvements.

Comments on 'Current GCMs' Unrealistic Negative Feedback in the Arctic'

In contrast to prior studies showing a positive lapse-rate feedback associated with the Arctic inversion, Boé et al. reported that strong present-day Arctic temperature inversions are associated with stronger negative longwave feedbacks and thus reduced Arctic amplification in the model ensemble from phase 3 of the Coupled Model Intercomparison Project (CMIP3). A permutation test reveals that the relation between longwave feedbacks and inversion strength is an artifact of statistical self-correlation and that shortwave feedbacks have a stronger correlation with intermodel spread. The present comment concludes that the conventional understanding of a positive lapse-rate feedback associated with the Arctic inversion is consistent with the CMIP3 model ensemble.

Inference in differences-in-differences with few treated groups and heteroskedasticity

Differences-in-Differences (DID) is one of the most widely used identification strategies in applied economics. However, how to draw inferences in DID models when there are few treated groups remains an open question. We show that the usual inference methods used in DID models might not perform well when there are few treated groups and errors are heteroskedastic. In particular, we show that when there is variation in the number of observations per group, inference methods designed to work when there are few treated groups tend to (under-) over-reject the null hypothesis when the treated groups are (large) small relative to the control groups. This happens because larger groups tend to have lower variance, generating heteroskedasticity in the group x time aggregate DID model. We provide evidence from Monte Carlo simulations and from placebo DID regressions with the American Community Survey (ACS) and the Current Population Survey (CPS) datasets to show that this problem is relevant even in datasets with large numbers of observations per group. We then derive an alternative inference method that provides accurate hypothesis testing in situations where there are few treated groups (or even just one) and many control groups in the presence of heteroskedasticity. Our method assumes that we can model the heteroskedasticity of a linear combination of the errors. We show that this assumption can be satisfied without imposing strong assumptions on the errors in common DID applications. With many pre-treatment periods, we show that this assumption can be relaxed. Instead, we provide an alternative inference method that relies on strict stationarity and ergodicity of the time series. Finally, we consider two recent alternatives to DID when there are many pre-treatment periods. We extend our inference methods to linear factor models when there are few treated groups. We also derive conditions under which a permutation test for the synthetic control estimator proposed by Abadie et al. (2010) is robust to heteroskedasticity and propose a modification on the test statistic that provided a better heteroskedasticity correction in our simulations.

Inference in differences-in-differences with few treated groups and heteroskedasticity

Differences-in-Differences (DID) is one of the most widely used identification strategies in applied economics. However, how to draw inferences in DID models when there are few treated groups remains an open question. We show that the usual inference methods used in DID models might not perform well when there are few treated groups and errors are heteroskedastic. In particular, we show that when there is variation in the number of observations per group, inference methods designed to work when there are few treated groups tend to (under-) over-reject the null hypothesis when the treated groups are (large) small relative to the control groups. This happens because larger groups tend to have lower variance, generating heteroskedasticity in the group x time aggregate DID model. We provide evidence from Monte Carlo simulations and from placebo DID regressions with the American Community Survey (ACS) and the Current Population Survey (CPS) datasets to show that this problem is relevant even in datasets with large numbers of observations per group. We then derive an alternative inference method that provides accurate hypothesis testing in situations where there are few treated groups (or even just one) and many control groups in the presence of heteroskedasticity. Our method assumes that we know how the heteroskedasticity is generated, which is the case when it is generated by variation in the number of observations per group. With many pre-treatment periods, we show that this assumption can be relaxed. Instead, we provide an alternative application of our method that relies on assumptions about stationarity and convergence of the moments of the time series. Finally, we consider two recent alternatives to DID when there are many pre-treatment groups. We extend our inference method to linear factor models when there are few treated groups. We also propose a permutation test for the synthetic control estimator that provided a better heteroskedasticity correction in our simulations than the test suggested by Abadie et al. (2010).

The use of nonparametric contrasts in one-way layouts and random block designs

Nonparametric simple-contrast estimates for one-way layouts based on Hodges-Lehmann estimators for two samples and confidence intervals for all contrasts involving only two treatments are found in the literature.Tests for such contrasts are performed from the distribution of the maximum of the rank sum between two treatments. For random block designs, simple contrast estimates based on Hodges-Lehmann estimators for one sample are presented. However, discussions concerning the significance levels of more complex contrast tests in nonparametric statistics are not well outlined.This work aims at presenting a methodology to obtain p-values for any contrast types based on the construction of the permutations required by each design model using a C-language program for each design type. For small samples, all possible treatment configurations are performed in order to obtain the desired p-value. For large samples, a fixed number of random configurations are used. The program prompts the input of contrast coefficients, but does not assume the existence or orthogonality among them.In orthogonal contrasts, the decomposition of the value of the suitable statistic for each case is performed and it is observed that the same procedure used in the parametric analysis of variance can be applied in the nonparametric case, that is, each of the orthogonal contrasts has a chi(2) distribution with one degree of freedom. Also, the similarities between the p-values obtained for nonparametric contrasts and those obtained through approximations suggested in the literature are discussed.

Técnicas de análise de dados distribuídos em áreas

Pós-graduação em Biometria - IBB

Understanding the impact of Twitter on the financial market

Negli anni la funzione dei social network è cambiata molte volte. Alle origini i social network erano uno strumento di connessione tra amici, ora sono siti internet in cui le persone mettono informazioni e quando un social network ha milioni di utenti, diventa un’incredibile sorgente di dati. Twitter è uno dei siti internet più visitati, e viene descritto come “the SMS of internet”, perchè è un social network che permette ai suoi utenti di inviare e leggere messaggi corti, di 140 caratteri, chiamati “tweets”. Con il passare del tempo Twitter `e diventato una fonte fondamentale di notizie. Il suo grande numero di utenti permette alle notizie di espandersi nella rete in modo virale. Molte persone hanno cercato di analizzare il potere dei tweet, come il contenuto positivo o negativo, mentre altri hanno cercato di capire se avessero un potere predittivo. In particolare nel mondo finanziario, sono state avviate molte ricerche per verificare l’esistenza di una effettiva correlazione tra i tweets e la fluttuazione del mercato azionario. L’effettiva presenza di tale relazione unita a un modello predittivo, potrebbe portare allo sviluppo di un modello che analizzando i tweets presenti nella rete, relativi a un titolo azionario, dia informazioni sulle future variazioni del titolo stesso. La nostra attenzione si è rivolata alla ricerca e validazione statistica di tale correlazione. Sono stati effettuati test su singole azioni, sulla base dei dati disponibili, poi estesi a tutto il dataset per vedere la tendenza generale e attribuire maggior valore al risultato. Questa ricerca è caratterizzata dal suo dataset di tweet che analizza un periodo di oltre 2 anni, uno dei periodi più lunghi mai analizzati. Si è cercato di fornire maggior valore ai risultati trovati tramite l’utilizzo di validazioni statistiche, come il “permutation test”, per validare la relazione tra tweets di un titolo con i relativi valori azionari, la rimozione di una percentuale di eventi importanti, per mostrare la dipendenza o indipendenza dei dati dagli eventi più evidenti dell’anno e il “granger causality test”, per capire la direzione di una previsione tra serie. Sono stati effettuati anche test con risultati fallimentari, dai quali si sono ricavate le direzioni per i futuri sviluppi di questa ricerca.