Banda de Confiança

Visão geral

Bandas de confiança visualizam incerteza mostrando um intervalo ao redor de uma linha de tendência central. A banda representa valores possíveis dentro de um intervalo de confiança (por exemplo, IC 95%), ajudando o público a entender a precisão de estimativas ou previsões.

Quando usar

• Mostrar incerteza de previsão em modelos de forecasting
• Exibir intervalos de confiança estatísticos
• Visualizar margens de erro em medições
• Apresentar a faixa de resultados possíveis
• Comunicar incerteza de modelos em ciência de dados

Não é ideal

• Quando a incerteza é desprezível ou irrelevante
• Com públicos não familiarizados com conceitos estatísticos
• Quando limites exatos são mais importantes do que tendências

Variações principais

• Um nível de confiança vs múltiplos (por exemplo, 50%, 90%, 95%)
• Bandas simétricas vs assimétricas
• Largura de confiança constante vs variando no tempo
• Combinado com distinção entre valor real e previsto

Casos de uso

• Previsão do tempo: faixas de temperatura previstas
• Projeções financeiras: previsão de receita com limites de confiança
• Aprendizado de máquina: previsões do modelo com incerteza
• Ensaios clínicos: estimativas de efeito com intervalos de confiança
• Previsão de vendas: cenários de limite superior e inferior

Dados (CSV)

date,Lower,Upper,Value
2012-08-28,-0.282,0.240,-0.162
2012-08-29,-0.042,0.060,-0.012
2012-08-30,-0.027,0.060,0.003
2012-08-31,-0.040,0.060,-0.010
2012-09-01,-0.038,0.060,-0.008
2012-09-02,-0.034,0.060,-0.004
2012-09-03,-0.038,0.060,-0.008
2012-09-04,-0.035,0.060,-0.005
2012-09-05,-0.039,0.060,-0.009
2012-09-06,-0.037,0.060,-0.007
2012-09-07,-0.035,0.060,-0.005
2012-09-08,-0.030,0.060,0.000
2012-09-09,-0.037,0.060,-0.007
2012-09-10,-0.033,0.060,-0.003
2012-09-11,-0.031,0.060,-0.001
2012-09-12,-0.030,0.060,0.000
2012-09-13,-0.033,0.060,-0.003
2012-09-14,-0.032,0.060,-0.002
2012-09-15,-0.029,0.060,0.001
2012-09-16,-0.027,0.060,0.003
2012-09-17,-0.021,0.060,0.009
2012-09-18,-0.052,0.060,-0.022
2012-09-19,-0.039,0.060,-0.009
2012-09-20,-0.030,0.060,0.000
2012-09-21,0.019,0.040,0.039
2012-09-22,-0.017,0.060,0.013
2012-09-23,-0.030,0.060,0.000
2012-09-24,-0.036,0.060,-0.006
2012-09-25,-0.036,0.060,-0.006
2012-09-26,-0.036,0.060,-0.006
2012-09-27,-0.034,0.060,-0.004
2012-09-28,-0.043,0.060,-0.013
2012-09-29,-0.039,0.060,-0.009
2012-09-30,-0.039,0.060,-0.009
2012-10-01,-0.043,0.060,-0.013
2012-10-02,-0.041,0.060,-0.011
2012-10-03,-0.041,0.060,-0.011
2012-10-04,-0.041,0.060,-0.011
2012-10-05,-0.040,0.060,-0.010
2012-10-06,-0.037,0.060,-0.007
2012-10-07,-0.034,0.060,-0.004
2012-10-08,-0.032,0.060,-0.002
2012-10-09,0.047,0.060,0.077
2012-10-10,0.094,0.060,0.124
2012-10-11,0.080,0.060,0.110
2012-10-12,0.057,0.060,0.087
2012-10-13,0.038,0.060,0.068
2012-10-14,0.030,0.060,0.060
2012-10-15,-0.042,0.060,-0.012
2012-10-16,-0.051,0.060,-0.021
2012-10-17,-0.042,0.060,-0.012
2012-10-18,-0.040,0.060,-0.010
2012-10-19,-0.038,0.060,-0.008
2012-10-20,-0.037,0.060,-0.007
2012-10-21,-0.034,0.060,-0.004
2012-10-22,-0.034,0.060,-0.004
2012-10-23,-0.036,0.060,-0.006
2012-10-24,-0.035,0.060,-0.005
2012-10-25,-0.035,0.060,-0.005
2012-10-26,-0.035,0.060,-0.005
2012-10-27,-0.034,0.060,-0.004
2012-10-28,-0.031,0.060,-0.001
2012-10-29,-0.033,0.060,-0.003
2012-10-30,-0.031,0.060,-0.001
2012-10-31,-0.030,0.060,0.000
2012-11-01,-0.032,0.060,-0.002
2012-11-02,-0.032,0.060,-0.002
2012-11-03,-0.030,0.060,0.000
2012-11-04,-0.031,0.060,-0.001
2012-11-05,-0.029,0.060,0.001
2012-11-06,-0.028,0.060,0.002
2012-11-07,-0.029,0.060,0.001
2012-11-08,-0.029,0.060,0.001
2012-11-09,-0.027,0.060,0.003
2012-11-10,-0.028,0.060,0.002
2012-11-11,-0.028,0.060,0.002
2012-11-12,-0.026,0.060,0.004
2012-11-13,-0.027,0.060,0.003
2012-11-14,-0.027,0.060,0.003
2012-11-15,-0.026,0.060,0.004
2012-11-16,-0.027,0.060,0.003
2012-11-17,-0.025,0.060,0.005
2012-11-18,-0.025,0.060,0.005
2012-11-19,-0.024,0.060,0.006
2012-11-20,-0.023,0.060,0.007
2012-11-21,-0.025,0.060,0.005
2012-11-22,-0.025,0.060,0.005
2012-11-23,-0.023,0.060,0.007
2012-11-24,-0.022,0.060,0.008
2012-11-25,-0.022,0.060,0.008
2012-11-26,0.004,0.060,0.034
2012-11-27,0.029,0.060,0.059
2012-11-28,0.049,0.060,0.079

Configuração do Gráfico (JSON)

O gráfico cria uma banda de confiança empilhando duas linhas invisíveis (limites Lower e Upper) para formar uma área sombreada, com a linha Value exibida de forma destacada por cima.

Nota importante sobre o formato dos dados: para que a técnica de empilhamento funcione corretamente, a coluna "Upper" no CSV deve conter a diferença entre o limite superior e o limite inferior (Upper - Lower), e não o valor absoluto do limite superior. Isso acontece porque o ECharts empilha valores de forma aditiva: Lower + (Upper - Lower) = limite superior real. Por exemplo:

• Se o seu limite inferior é -0.282 e o limite superior é -0.042
• O CSV deve ter: Lower = -0.282, Upper = 0.240 (que é -0.042 - (-0.282))

Dicas de desempenho

• Use preenchimento de área com baixa opacidade para a banda
• Rotule claramente o nível de confiança
• Considere ocultar a banda ao passar o mouse para revelar valores

Perguntas frequentes

O que as diferentes linhas representam? Em um gráfico de banda de confiança:

• A linha central mostra sua melhor estimativa ou previsão
• O limite superior representa o valor máximo esperado em um determinado nível de confiança
• O limite inferior representa o valor mínimo esperado
• A área sombreada entre limites superior e inferior visualiza a faixa de incerteza

Qual nível de confiança devo usar? 95% é o mais comum em contextos científicos. 90% pode ser usado em análises preliminares, enquanto 99% indica maior certeza. Sempre indique qual nível você está exibindo.

Como explicar bandas de confiança para públicos não técnicos? Use linguagem simples, por exemplo:

• "A área cinza mostra onde esperamos que o valor real esteja"
• "A região sombreada representa nossa incerteza"
• "A linha é nossa melhor estimativa; a banda mostra o intervalo de possibilidades"

Evite jargões como "intervalo de confiança" ou "erro padrão" a menos que o público seja técnico.

Por que a largura da banda muda ao longo do tempo? A largura reflete mudanças no nível de incerteza. Razões comuns:

• Menos dados → bandas mais largas (mais incerteza)
• Mais volatilidade → bandas mais largas
• Início/fim de séries temporais frequentemente têm bandas mais largas
• Eventos ou intervenções podem aumentar temporariamente a incerteza

Devo mostrar múltiplos níveis de confiança? Em geral, não. Múltiplas bandas (por exemplo, 50%, 90%, 95%) podem confundir públicos não técnicos. Use um nível alinhado ao padrão do seu campo. Em artigos acadêmicos, você pode mostrar 95% no gráfico principal e 90%/99% em anexos.

Como lidar com intervalos de confiança assimétricos? Use colunas separadas para limites superior e inferior. Bandas assimétricas são comuns em:

• Distribuições não normais
• Dados de contagem (modelos de Poisson)
• Proporções próximas de 0 ou 1
• Dados com transformação log

Que cor devo usar para a banda de confiança? Cinza é a escolha mais comum e neutra: não sugere julgamentos de bom/ruim. Se precisar seguir a paleta da marca, use um tom bem claro com baixa opacidade (~20–30%) para não competir com a linha principal.